Professional Documents
Culture Documents
of (Coiriificss
MATHEMATICAL W 0
(Geco»nnicetry
of (Comiikcss
AV V. AVhecoylpyauvn
A. A. Zassllanvssky
AmEw
AMS
www.ams.org
I'EOMETPI4gECKME CBOIICTBA KPMBbIX BTOPOI'O IIOPSI.IIKA
A. B. AxOruIH, A. A. 3acJIaBcxHH
MIIHMO, MOCKBA, 2007
This work was originally published in Russian by Moscow Center for Continuous
Mathematical Education under the title "I'eoMeTpHxeecaHe caoticTaa xp1BbIx sToporo
nopii,z a" © 2007. The present translation was created under license for the American
Mathematical Society and is published by permission.
Copying and reprinting. Individual readers of this publication, and nonprofit libraries
acting for them, are permitted to make fair use of the material, such as to copy a chapter for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for such
permission should be addressed to the Acquisitions Department, American Mathematical Society,
201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by
e-mail to reprint -permission@ams . org.
© 2007 by the American Mathematical Society. All rights reserved.
The American Mathematical Society retains all rights
except those granted to the United States Governpnent.
Printed in the United States of America.
® The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at http://www.ams.org/
10987654321 121110090807
Contents
Preface vii
Chapter 1. Elementary Properties of Curves of Second Degree 1
1.1. Definitions 1
1.2. Analytic definition and classification of curves
of second degree 5
1.3. The optical property 6
1.4. The isogonal property of conics 10
1.5. Curves of second degree as projections of the circle 15
1.6. The eccentricity and yet another definition of conics 17
1.7. Some remarkable properties of the parabola 19
V
vi CONTENTS
Bibliography 131
Index 133
Preface
Curves of second degree, or conics, are traditionally viewed as objects per-
taining to analytic geometry and are studied in lower-level courses in engi-
neering colleges. At best, only the optical properties of conics are mentioned
among their geometric properties. But those curves also possess a number
of other nice properties, a majority of which can be established by meth-
ods of elementary geometry well within the reach of high school students.
Moreover, conics help solve some geometric problems seemingly unrelated
to conics. In this book the reader will find the most interesting facts about
curves of order two, including those proved recently.
Chapter 1 deals with the elementary properties of conics. Most of the
facts mentioned there are well known. The remaining material is also rather
simple, so that the entire chapter does not impose any prerequisites on
the reader beyond the standard high school curriculum. Some simple but
important results are offered as exercises. We recommend that the reader
try to solve them before reaching for the solutions. This should facilitate the
understanding of the material later on. Chapter 2 is of an auxiliary nature.
It contains some facts from classical geometry needed for understanding
the remaining chapters, which are not usually studied in high school. In
Chapter 3 we mention projective properties common to all conics. Some
of them, such as the theorem on pencils of conics, are quite complicated.
Finally, Chapter 4 is devoted to metric properties. As a rule, they concern
only special kinds of conics. This is the most complicated chapter of the
book, which requires a good understanding of the material in the previous
chapters.
The authors are grateful to I. I. Bogdanov and E. Yu. Bun'kova for
valuable comments.
Vii
Chapter 1
Elementary Properties of
Curves of Second Degree
1.1. Definitions
If you stake a goat, it will graze the grass inside the circle that is centered
at the stake and has radius the length of the rope. If you use two stakes at
the ends of the rope and tie the goat using a sliding ring, the region with
grazed grass will look like the one shown in Figure 1.1.
FIGURE 1.1. F1 and F2 are the foci; a and b are the major
and the minor axes.
For all points on the boundary of that figure, the sum of the distances
to the stakes equals the length of the rope. Such a curve is called an ellipse,
and the points marked by the stakes are called the foci.
Clearly, an ellipse looks like an "elongated circle". It obviously has two
axes of symmetry. These are the line connecting the foci and the midpoint
perpendicular to the segment with endpoints at the foci. These two lines
are called the major and the minor axes of the ellipse. The lengths of their
parts inside the ellipse are called the lengths of the major and minor axes.
The distance between the foci is called the focal distance.
It is also clear that the length of the rope holding the goat equals the
length of the major axis of the elliptical boundary of the grazed region.
1
2 1. ELEMENTARY PROPERTIES OF CURVES OF SECOND DEGREE
Intuitively it is clear that the goat can graze at any point inside the
ellipse but it can never get beyond the ellipse. But a purely mathematical
reformulation of this is no longer so obvious.
Exercise 1. Prove that the sum of the distances from any point inside the
ellipse to the foci is less-and from any point outside the ellipse is greater-
than the length of the major axis.
Solution. Denote by F1 and F2 the foci of the ellipse, and by X a point.
Let Y be the intersection of the ray F1X and the ellipse. Assume first that
X is inside the ellipse. By the triangle inequality, F2X < XY + YF2, and
hence FIX + XF2 < F1X + XY + YF2 = F1Y + F2Y (Figure 1.2).
FIGURE 1.2
But F1Y + F2Y equals the length of the rope, i.e., the major axis of
the ellipse. Using a similar argument when X is outside the ellipse, we have
F2Y < XY+XF2. Therefore FIX+XF2 = FIY+YX+XF2 > FiY+F2Y.
Ellipses often arise in mechanics. For example, a planet orbiting the Sun
moves along an ellipse with the Sun at one of its foci (Kepler's Law).
An ellipse is an example of a curve of second degree or a conic. Other
examples of such curves are parabolas and hyperbolas.
A hyperbola is the set of points for which the absolute value of the differ-
ence between the distances to two fixed points, called the foci, is constant.
A hyperbola consists of two branches the ends of which approach two
lines called the asymptotes of the hyperbola (Figure 1.3). A hyperbola with
perpendicular asymptotes is said to be equilateral.
The line passing through the foci of a hyperbola is an axis of symmetry
and is called the real axis. The perpendicular line passing through the
midpoint between the foci is also an axis of symmetry and is called the
imaginary axis of the hyperbola.
If a comet is passing by the Sun and the gravitational force exerted by
the Sun is too small to keep the comet within the solar system, then its
trajectory will be an arc of a hyperbola whose focus will be at the center of
the Sun.
A parabola is the set of points whose distances to some fixed point and
line are constant. That point and line afe called, respectively, the focus and
the directrix of the parabola. The line perpendicular to the directrix and
passing through the focus is called the axis of the parabola (Figure 1.4).
1.1. DEFINITIONS 3
FIGURE 1.3. Fl and F2 are the foci, a and b are the real and
imaginary axes, and li and 12 are the asymptotes.
FIGURE 1.4. F is the focus; l and 1' are the directrix and the
axis of the parabola.
FIGURE 1.5
is obvious. Suppose X' is outside the parabola but on the same side of
the directrix. Then Z'Y' = Z'X' + X'Y' and, by the triangle inequality,
FX' + X'Z' > FZ' = Z'Y' = Z'X' + X'Y'. Therefore FX' > X'Y'.
In the case of a hyperbola the corresponding statement is as follows: let
d be the difference of the distances from any point on the hyperbola to the
foci F1 and F2 and let IF be the branch of the hyperbola inside which F1
lies. Then for the points X outside (inside) F the quantity XF2 - XF1 is
less (greater) than d.
Suppose X lies inside F and let Y be the intersection of the ray F2X and
F. We have F2X = F2Y+YX. By the triangle inequality, F1X < F1Y+YX;
therefore F2X - FIX > (F2Y + YX) - (F1Y + YX) = F2Y - F1Y = d.
FIGURE 1.6
The curve
x2
(II) = 1, a>0, b > 0,
a2 b2
is a hyperbola that intersects its real axis in two points at distance 2a from
each other. The quantities a and b are called, respectively, the real and
the imaginary semi-axes of the hyperbola. The lines x/y = ±a/b are the
asymptotes of the hyperbola and the points -+b2, 0) are the foci.
When a = b hyperbola (II) is equilateral.
If
(III) y2 = 2px, P>0'
the curve is a parabola, whose axis coincides with the x-axis, the focus is at
(p/2, 0), and the directrix is given by x = -p/2.
The curve
x2 2
a2+b2 =-1
is called an imaginary ellipse; it contains no real points.
Henceforth, unless stated otherwise, a curve of degree two will always
be nondegenerate and not imaginary.
Problem 1. Prove that the equation y = 1/x describes a hyperbola and
find its foci.
FIGURE 1.7
IF1P - F2PI < F1F2 and the maximum is attained if and only if F1, F2 and
P lie on a straight line. Since the points F2 and F2 are the reflections of
each other, the angles formed by the lines F1P and F2P with 1 are equal
(Figure 1.8).
F1
FIGURE 1.8
b) Let F' be the reflection of F in 1. Of the two points F and F' choose
the one whose (signed) distance to 1' is minimal. Let it be F and let d be
the distance from F to 1'. Then for any point P on 1 the distance to 1' is
not greater than PF + d. Therefore the difference in question never exceeds
d. On the other hand, it is exactly d when P lies on the perpendicular to 1'
passing through F (Figure 1.9).
We also note that if the line F1F2 in a) is parallel to 1 and the line 1' in b)
is perpendicular to 1, then there is no maximum (it is attained at infinity).
Now we state one of the most important properties of conics, the so-
called optical property.
FIGURE 1.9
8 1. ELEMENTARY PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 1.10
FIGURE 1.11
FIGURE 1.12
FIGURE 1.13
10 1. ELEMENTARY PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 1.14
Proof. By the optical property, PR and QR are the bisectors of the exterior
angles of the triangle F2PQ. Therefore R is the center of an excircle. The
tangency point (call it Fl) of the excircle and the corresponding side and the
point F2 cut the perimeter of the triangle into equal parts, i.e., F1P+PF2 =
F2Q+QFF. But F1 has this property and there is only one such point. Hence
Fl' and F1 coincide.
Corollary. The straight line connecting a focus of an ellipse and the inter-
section of the tangents to the ellipse at the ends of a chord containing that
focus is perpendicular to the chord.
For the hyperbola, Theorem 1.2 is also true but the excircle should be
replaced by the incircle.
FIGURE 1.15
Thus, the triangles PF2Fi and PF1F2 are equal (having three equal sides).
Therefore
LF2PF1 + 2LF1PX = LF2PFi = LF1PF2 = LF1PF2 + 2LF2PY.
Hence LF1PX = LF2PY, which is the desired result.1
Figure 1.16 shows that a similar property holds for the hyperbola.2
FIGURE 1.16
Suppose now that the ellipse (or hyperbola) with foci F1 and F2 is in-
scribed in triangle ABC. It follows from the above that LBAF1 = LCAF2i
LABF1 = LCBF2 and LACF1 = LBCF2.
We shall show in 2.3 that, in a plane, for any (with rare exceptions)
point X there is a unique point Y such that X and Y are the foci of a
'We consider the case when F1 and F2 are inside the angle F1PF2' and F1 lies inside
the angle F2PFi. In the remaining cases the arguments are similar.
2The reader should check two cases: when the tangency points are either on different
branches or on the same branch.
12 1. ELEMENTARY PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 1.17
Theorem 1.5. The locus of points from which a given ellipse is seen at
a right angle (i.e., the tangents to the ellipse drawn from such a point are
perpendicular) is a circle centered at the center of the ellipse (Figure 1.18).
FIGURE 1.18
1.4. THE ISOGONAL PROPERTY OF CONICS 13
Proof. Let Fl and F2 be the foci of the ellipse and suppose that the tangents
to the ellipse at X and Y intersect in P. Reflecting F1 in PX we have a
point Fi. It follows from Theorem 1.3 that LXPY = LF1PF2 and FFF2 =
F1X +F2X, i.e., the length of the segment F1'F2 equals the major axis of the
ellipse (the length of the rope tying the goat). The angle F1PF2 is right if
and only if F1P2 + F2P2 = FiF2 (by the Pythagorean theorem). Therefore
XPY is a right angle if and only if F1P2 + F2P2 equals the square of the
major axis of the ellipse. But it is not difficult to see that this condition
defines a circle. Indeed, suppose F1 has Cartesian coordinates (x1i yi), and
F2 has coordinates (x2i y2). Then the coordinates of the desired points P
satisfy the condition
(x - xl)2 + (y - yl)2 + (x - x2)2 + (y - y2)2 = C,
where C is the square of the major axis. But since the coefficients of x2
and y2 are equal (to 2) and the coefficient of xy is zero, the set of points
satisfying this condition is a circle. By virtue of symmetry, its center is the
midpoint of the segment F1F2.
For the hyperbola such a circle does not always exist. When the angle
between the asymptotes of the hyperbola is acute, the radius of the circle is
imaginary. If the asymptotes are perpendicular, then the circle degenerates
into the point which is the center of the hyperbola.
Example. Given points P1,. .. , P,,, and numbers k1,. .. , kn, and C, the locus
of points X such that k1XP1 + + k,,,XPn = C is a circle, known as the
Fermat-Apollonius circle. Clearly, it may have an imaginary radius (when?).
Theorem 1.6. Suppose a string is put on an ellipse a and then pulled
tight using a pencil. If the pencil is rotated about the ellipse, it will traverse
another ellipse confocal with a (Figure 1.19).
FIGURE 1.19
14 1. ELEMENTARY PROPERTIES OF CURVES OF SECOND DEGREE
Proof. Clearly, the new figure (call it al) has a smooth boundary. We shall
show that at each point X on al the tangent to the new curve coincides with
the bisector of the exterior angle F1XF2.
Let XM and XN be the tangents to a. Then LF1XN = LF2XM, and
hence the bisector l of the exterior angle NXM coincides with the bisector
of the exterior angle F1 X F2. Call it 1.
Let Y be an arbitrary point on 1 and YL and YR the tangents to a, as
shown in Figure 1.19. We assume that Y lies "to the left" of X; the other
case is argued similarly.
Let P be the intersection of the lines XM and YL. It is easy to see that
YN < YR+ -RN, and -LM < LP+PM. Moreover, since 1 is the exterior
bisector of the angle NXP, we have PX + X N < PY + YN. Therefore
MX +XN+-NM < MX +XN+vNL+LP+PM
= PX +XN+vNL+LP < PY+YN+vNL+LP
= LY+YN+'-NL
< LY+YR+vRN+vNL = LY+YR+-RL
(here the arcs are meant to be the arcs under the string). Therefore Y lies
outside a1. The same is true for any point Y on 1. It follows that a1 contains
a single point of 1, i.e., the line is tangent. It also follows at once that the
obtained curve is convex.
Thus the sum of the distances to the foci F1 and F2 does not change
with time. Therefore the trajectory of the pencil is an ellipse.
Here is a more rigorous approach to the last claim. Suppose X is outside the
ellipse. Put the pencil at X and pull the string around it and around the ellipse.
Let f (X) be the length of the string and g(X) = F1X +F2X (a point is understood
as a pair of its coordinates; thus both f and g depend on a pair of real numbers).
One can show that those functions are continuously differentiable and that the
vectors grad f = (a , and grad g = (a , a2) are nonzero at each point. Then,
by the implicit function theorem, the curve traversed by the pencil with a string of
fixed length (i.e., a level curve of f) is smooth (continuously differentiable). It now
follows that the curve can be parametrized by a differentiable function R = R(t)
(this is again a pair of coordinate functions x = x(t), y = y(t)) whose tangent
vector is different from zero. As shown before, the tangent vector dft = (at de )
of the curve is tangent to a level curve of g, i.e., it is perpendicular to gradg(R) at
R = R(t). Consider the function g(R(t)). Its derivative is
dg(R(t)) 8g dx(t) i9g dy(t)
0
dt ex dt + 8y dt
(this is the orthogonality condition mentioned above), i.e., g(R(t)) is constant. This
means that our curve lies on an ellipse with the same foci. Since any ray starting
at F1 must contain a point on our curve, the curve coincides with the ellipse.
FIGURE 1.20
FIGURE 1.21
Finally, consider the case when the secant plane is parallel to one ruling
(Figure 1.22).
FIGURE 1.22
intersection of the cone and the plane ir. Let Y be the intersection of the
line SX with a and Z the projection of X to 1. We shall show that the ratio
of XY and XZ is constant, i.e., does not depend on X.
FIGURE 1.23
FIGURE 1.24
Proof. Suppose a line l touches the parabola at P and let P' be the pro-
jection of P to the directrix. Since the triangle FPP' is isosceles and l is
the bisector of the angle P, l is an axis of symmetry of the triangle. Hence
the reflection P of F in 1 is on the directrix.
Corollary. The projections of the focus of the parabola to its tangents lie
on the line tangent to the parabola at its vertex. (Figure 1.25).
FIGURE 1.25
Lemma 1.2. Suppose the tangents to the parabola at points X and Y in-
tersect at a point P. Then P is the center of the circumcircle of the triangle
FX'Y', where X' and Y' are the projections of X and Y to the directrix of
the parabola, and F is the focus of the parabola (Figure 1.26).
FIGURE 1.26
FIGURE 1.27
FIGURE 1.28
Proof. Suppose P lies on the directrix, and let X' and Y' be the projections
of X and Y to the directrix. Then the triangles PXF and PXX' are
equal (since they are symmetric with respect to PX). Hence LPFX =
LPX'X = 90°. Similarly, LPFY = LPY'Y = 90°. Moreover, LXPY =
2 (LFPX' + LFPY') = 90°. The fact that there are no other points with
this property is obvious.
1.7. SOME REMARKABLE PROPERTIES OF THE PARABOLA 21
Since similar assertions are true for the remaining conics, the above
theorem seems to be rather natural. However, the first part of the theorem
has an unexpected generalization that holds only for parabolas. It will be
used later in 3.2 in the proof of Fregier's theorem.
Theorem 1.8. The set of points from which a parabola is seen at an angle
0 or 180° - 0 is a hyperbola with focus F and directrix l (Figure 1.29).
FIGURE 1.29
FIGURE 1.30
and 1 is equal to the angle between Y'F and X'Y' because 1 is perpendicular
to X'Y' (the directrix of the parabola) and Y'F is perpendicular to PY
(moreover, PY is the midpoint perpendicular to Y'F). This proves the first
part of the theorem.
We now prove the second part. Since 1 is parallel to YY', the angle
between PY and l is equal to the angle PYY', which, by the optical property,
is equal to the angle FYP. Thus LFYP = LX PF. Similarly, LFX P =
LYPF. Therefore the triangles X FP and PFY are similar.
The next theorem is actually a consequence of Theorem 1.9. But we shall
prove it using Simson's line, which will help us find even more interesting
properties of the parabola.
Theorem 1.10. Suppose a triangle ABC is circumscribed about a parabola
(i.e., the lines AB, BC, CA are tangent to the parabola). Then the focus
of the parabola lies on the circumcircle of the triangle ABC.
Proof. By the Corollary of Lemma 1.1, the projections of the focus to the
sides all lie on a straight line (which is parallel to the directrix and lies at
half the distance from the focus). Now we can use Simson's lemma.
Lemma 1.3 (Simson). The projections of P to the sides of a triangle ABC
lie on a line if and only if P lies on the circumcircle of the triangle.
Proof. Let Pa, Pb and Pc be the projections of P to BC, CA and AB,
respectively. We consider the case shown in Figure 1.31; the remaining
cases are argued similarly.
The quadrilateral PPPbPa is inscribed, hence LPPbPa = LPCPa. Sim-
ilarly, LPPbPc = LPAPC. The points Pa, Pb and Pc lie on a line if and only
if LPPbPc = LPPbPa or, equivalently, LPAPC = LPCPa. But this means
that P lies on the circumcircle of the triangle ABC. The remaining cases
are argued similarly.
An identical argument proves the converse. If P lies on the circumcircle
of a triangle ABC, then LPAB = LPCPa = LPPbPa (the latter holds since
P, C, Pa and Pb lie on a circle). Similarly, LPAB = LPPbPc. Therefore
Pa, Pb and Pc lie on a straight line.
1.7. SOME REMARKABLE PROPERTIES OF THE PARABOLA 23
FIGURE 1.31
FIGURE 1.32
FIGURE 1.33
Proof. Consider the case shown in Figure 1.33; the remaining cases are
argued similarly. Let PP and Pb be the projections of P to the sides AB
and AC, respectively. Then LABB' = LAPB' as the angles subtending the
arc AB'. Since quadrilateral APCPbP is inscribed (AP is a diameter of its
circumcircle) and the sum of the opposite angles of an inscribed quadrilateral
equals 180°, we have LAPB' = LAPPb = 180° - LAPCPb = LBPPb.
Therefore PbPc is parallel to BB'.
Corollary 1. When the point P moves along the circle, Simson's line rotates
in the opposite direction with velocity one half the rate of change of the arc
PA.
Corollary 2. Simson's line of P relative to a triangle ABC cuts the segment
PH (where H is the orthocenter of the triangle ABC) in half (Figure 1.34).
FIGURE 1.34
Proof. It is easy to see that ZAHC = 180° - LABC, and therefore the
reflection H' of H in AC lies on the circumcircle of the triangle ABC.
Since the lines PB' and BH' are perpendicular to AC, the quadrilateral
PB'BH' is a trapezoid; being inscribed, it must be equilateral. Therefore
1.7. SOME REMARKABLE PROPERTIES OF THE PARABOLA 25
FIGURE 1.35
FIGURE 2.1
FIGURE 2.2
The Euler circle or the nine-point circle is the circle passing through the
midpoints of the sides of a triangle ABC. It turns out that it also intersects
the sides at the feet of the heights. Moreover, if H denotes the orthocenter
of the triangle, then the midpoints of the segments AH, BH, and CH also
lie on that circle.
Now we shall prove it. Let Ma,, Mb, and MM be the midpoints of the
sides and Ha, Hb, and HH the feet of the heights (Figure 2.2). We show that
the angles MbMaMC and MbHaMc are equal. This would imply that Ha lies
on the Euler circle.
ACHa is a right triangle and therefore MbHa = MbA. Similarly, McHa =
MBA. Since MbA = MaMc and MA = MaMb, triangles MbMaMc and
MbHaMc are equal. Therefore the corresponding angles are also equal. Sim-
ilarly one shows that the points Hb and HH also lie on the Euler circle.
Notice that the feet of the heights of triangles ABC and ABH coincide
and therefore their Euler circles also coincide. Therefore the Euler circle
also contains the midpoints of the segments AH and BH. The fact that it
also contains the midpoint of the segment HC is established similarly.
Now we can prove Feuerbach's theorem.
Proof. Let Ga, Gb and G, be the tangency points of the incircle and the
sides of the triangle. Let Al be the foot of the bisector of the angle A and
C' the reflection of C in AA1. Let P be the intersection of AAI and W.
2.2. BASIC FACTS ABOUT PROJECTIVE TRANSFORMATIONS 29
FIGURE 2.3
Notice that P is the midpoint of CC' and therefore P lies on the midline
MaMb. Also notice that
Ma,P = (MaMb - PMbI = 2I AB -AC1 = 2I BGc - CGbl = 2 IBGa - CGaI
= MaGa.
Since the pairs of triangles MbMaC and ABC, ABA, and PMaAl are
similar, we have
MaP _BC'_MaQ
MaMb BA MaP'
where Q is the intersection of A,C' and MaMb. Hence, MaGa = MaP2 =
MaQ MaMb. Therefore the inversion with center Ma and radius MaGa
transforms Mb into a point lying on the line C'A,, the latter being the
reflection of BC in the bisector of the angle A. The same is true for M.
Thus the inversion with center Ma and radius MaGa transforms the Euler
circle into a line tangent to the incircle and therefore the nine-point circle
is also tangent to the incircle.
For excircles, Feuerbach's theorem is proved similarly. 0
The tangency point of the incircle and the Euler circle is called the
Feuerbach point (denoted F). Sometimes the tangency points of the excircles
and the Euler circle are also called the Feuerbach points (denoted Fa, Fb
and Fc).
at infinity, are viewed as the intersections of parallel lines and each point
at infinity is regarded as belonging to all the lines with the same direction.
The plane completed this way is called the projective plane.
FIGURE 2.4
FIGURE 2.5
FIGURE 2.6
FIGURE 2.7
Proof. Move the intersection of two diagonals to the center of the circle.
We need to show that the third diagonal passes through the center.
Thus let the hexagon ABCDEF be circumscribed about a circle cen-
tered at 0 and suppose that the diagonals AD and BE pass through 0. Let
Al, B1, ... , Fl be the tangency points of the circle and the sides AB, BC,...,
FA, respectively. It is easy to see that LE10C1 = LF1OB1 = 2LAOB and
that LE1OF = LFOF1 and LB1OC = LCOC1. Hence
3600
LFOF1+LF10B1+LB1OC = LE1OF+LEIOCI+LCOC1 =
2
Therefore F, 0 and C all lie on a straight line.
FIGURE 2.8
Notice that the theorems of Pascal and Brianchon remain true if the
hexagon degenerates into a pentagon or a quadrilateral. Later on, we will
use this observation on several occasions.
In the proof of Pascal's theorem we needed the assumption that the cor-
responding line does not intersect the circle, and in the proof of Brianchon's
theorem-that the intersection of the diagonals lies inside the circle. It turns
out that these two theorems are always true, i.e., the order of the points and
the lines in those theorems can be arbitrary.
2.2. BASIC FACTS ABOUT PROJECTIVE TRANSFORMATIONS 35
FIGURE 2.9
This implies that the pole of any line is the intersection of the polars of
all of its points and, dually, the polar of a point is the locus of the poles of
all the lines passing through the point.
2. The polar of a point A lying outside the circle will be the line con-
necting the tangency points of the circle and its tangents passing through
A (the tangency points are the poles of the corresponding tangents). It now
follows that, despite the metric properties mentioned in the definition, the
36 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.10
FIGURE 2.11
Problem 11. Given a circle, choose a point C inside (outside) it. Draw four
chords (secants) AiBi, i = 1, ... , 4 through C. Let D be the intersection
of the lines AlA2 and A3A4, and E the intersection of the lines B1B2 and
B3B4. Prove that C, D and E lie on a straight line (Figure 2.12).
FIGURE 2.12
FIGURE 2.13
Problem 14. The sides of a triangle ABC contain six points: A1, A2 on
BC, B1, B2 on AC and C1, C2 on AB. Prove that these six points lie on a
conic if and only if
BAI BA2 CB1 CB2 ACI AC2
CA1 CA2 AB1 AB2 BC1 BC2
Here the ratios are signed. For each expression, the positive direction is the
one from a vertex to the other vertex of the underlying side.
FIGURE 2.14
Ike)
em 2,B
ea¢l e9e ,e lee %
¢
.Alees ¢We f48C
2
°n e2jeu
a
(?hIS is st a refor
p
Sl so sle
resA) to ariaIlheslogle 14 C
De "-!?. eev¢ ngle o re
eth 0408 Ofhea2gie 1 a
BCO
the
c8¢
d
zzo Q'.ghe
CG'ptices'`Q
Ofp
ch ter
he tr'Iahglle
esAect o tcjr
or, eratr u f
/
FIG 2ZS
cet a tr1 slater 9t the
iSell
,
rtles Of O ent
the e2 C
Theee&2
¢n91e
be,B--it,
gl,g A
b Ce
he ve¢'
`rz'
ices
9le of
hick arAolnt o
s
th r Ae
r
ct to a
ft
d jhte echo s o e
.
c1 tr se
cie Ofthe la (Fib 216)
`4A with the rc
8,
16
Lei Tie e bee,, eo
eirC?
Zl¢r ¢n¢' 1'3e rienz
40 2. SOME RESULTS FROM CLASSICAL GEOMETRY
Proof. We consider the case shown in Figure 2.17. The remaining cases are
proved similarly.
The points PA, PB, PC are the vertices of the pedal triangle, and A',
B', C' are the vertices of the circumcevian triangle. We have LAA'C' =
LACC' = LPBPAP. The latter equality holds since the quadrilateral
PPACPB is inscribed. Similarly, LAA'B' = LPCPAP. Therefore LC'A'B'
= LPBPAPC Similarly, LA'B'C' = ZPAPBPC and LA'C'B' = LPAPCPB.
But this means that the triangles A'B'C' and PAPBPC are similar.
FIGURE 2.17
Proof. Since P and Q are inverse to each other, the triangles OAP and
OQA are similar (Figure 2.18), and therefore LOQA = LOAP. Since 0 is
the center of w, the triangle AOR is isosceles and therefore LOAR = LORA.
Thus
LPAR = LOAR - LOAP = LORA - LOQA = LRAQ,
as desired.
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 41
FIGURE 2.18
This lemma implies that w is the locus of points at which the ratio of
the distances to P and Q is constant (and different from 1). The circle w is
called the Apollonius circle of the segment PQ. We shall say more about it
later on.
Proof of the theorem. Suppose P and Q are inverse to each other with
respect to the circumcircle (call it w) of the triangle ABC and that PQ
intersects w in R. Let A' and A" be the second intersection points of AP
and AQ, respectively, with w. By Lemma 2.6, the line AR is the bisector of
the angle PAQ and therefore R cuts the arc A'A" in half, i.e., A' and A" are
symmetric with respect to OP. The same is true for the pairs B' and B",
C' and C", which are the vertices of the circumcevian triangles of P and Q
with respect to the triangle ABC. Thus the symmetry with respect to PQ
interchanges the triangles, and therefore they are similar and have opposite
orientations.
Corollary. The pedal triangles of inverse points are similar and have dif-
ferent orientations.
One can show that for any triangle XYZ there is a unique point such
that the pedal triangle of that point with respect to a given triangle ABC
is similar to the triangle XYZ with a fixed order of the vertices.
In the proof of the preceding theorem we had two similar triangles and a
point P inside them such that the angles formed by the sides and the Ceva
lines of P are equal but in a sense interchanged (see Figure 2.17).
Thus we have the so-called isogonal conjugation with respect to a trian-
gle.
Let ABC be an arbitrary triangle and P a point different from the
vertices of the triangle. Reflect the lines connecting the vertices of the
triangle with P in the bisectors of the corresponding angles of the triangle.
It turns out that these three lines always intersect in a single point (or are
parallel, i.e., intersect in a single point of the projective plane), which we
denote P' (Figure 2.19). The point P' is called the isogonal conjugate of P
with respect to triangle ABC; the transformation sending each point of the
projective plane to its isogonal conjugate is called the isogonal conjugation.
That the above concept is well defined was almost proved in Theorem 2.7.
Indeed, given a triangle ABC and a point P, let A'B'C' be the circumcevian
42 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.19
triangle of P with respect to the triangle ABC. Then ABC is the circum-
cevian triangle of P with respect to the triangle A'B'C' and therefore is
similar to the pedal triangle of P with respect to A'B'C'. Therefore the
image of P under the similarity transforming the pedal triangle of P into
the triangle ABC will be P, which is exactly the isogonal conjugate.
Now we mention several elementary properties of the isogonal conjuga-
tion.
1. If P does not lie on the sidelines of the triangle, then P' is determined
uniquely and the isogonal conjugate of P will be P. Such two points are
said to be isogonally conjugate.
2. The isogonal conjugate of a point on a sideline of the triangle is the
vertex of the triangle opposite to the respective side.
3. The isogonal conjugation leaves exactly four points of the plane
fixed-these are the centers of the incircle and of the three excircles of the
triangle.
4. If P lies on the circumcircle of the triangle ABC, then the isogonal
conjugate of P is the point on the line at infinity in the direction perpen-
dicular to Simson's line of P with respect to ABC (i.e., the line passing
through the projections of P to the sides of the triangle ABC).
FIGURE 2.20
The first three properties are obvious. To prove the fourth, consider the
case shown in Figure 2.20; the remaining cases are argued similarly. Suppose
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 43
FIGURE 2.21
Suppose a point P lies inside a triangle ABC and let Pa be its reflection
in the side BC; points Pb and P° are defined similarly (Figure 2.21). Let
P' be the center of the circumcircle of the triangle PaPbP°. The point
C is equidistant from Pa and Pb; therefore the line CP' is the midpoint
perpendicular to the segment PaPb. Hence LPaCP' = 2 LPaCPb = LC.
But then LBCP' = LPaCP' - LBCPa = LC - LBCP = LACP. A
similar argument shows that LABP' = LCBP and LBAP' = LCAP. But
this means that P' is isogonally conjugate to P with respect to ABC.
If P is outside the triangle, then the argument is identical, but if P
lies on the circumcircle of the triangle ABC, then the triangle PaPbPc is
degenerate. In this case the center of the circumcircle of the triangle PaPbPc
is not defined (although it would be natural to view the line PaPb as the
circumcircle, the center being the point on the line at infinity in the direction
perpendicular to PaPb).
The second method for constructing isogonally conjugate points shows
that the center of the pedal circle of P is the midpoint of the segment
PP' and its radius is half of the segment P'Pa, since the pedal circle of P
is the circle obtained from the circumcircle of the triangle PaPbP° by the
homothety with center P and coefficient 1 .
44 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.22
the points where the corresponding conics are hyperbolas, and in the clear
regions the conics are ellipses. The points of the circumcircle correspond to
parabolas.
FIGURE 2.23
If P and Q are isogonally conjugate in a polygon (i.e., for any vertex
X of this polygon the lines XP and XQ are symmetric with respect to the
bisector of the angle X), then there is a conic tangent to all the sides of the
polygon with foci at those points. The converse is also true, i.e., if a conic
is inscribed in a polygon, then its foci are isogonally conjugate with respect
to that polygon. Similarly one can show that in this case the pedal circles
of P and Q coincide (in addition to the fact that they exist!).
In the next section we shall show that for any five lines there is a unique
conic tangent to those lines. Hence a pentagon contains only one pair of
isogonally conjugate points. It is easy to see that for a quadrilateral such
points form a curve (actually, it will be a cubic, i.e., a curve of order three),
and for hexagons (and polygons with a larger number of sides) such points,
in general, do not exist.
Using isogonal conjugation one can easily prove Pascal's theorem in a
rather general form.
Theorem 2.9 (Pascal). Suppose points A, B, C, D, E and F lie on a
conic. Then the intersections of the lines AB and DE, BC and EF, CD
and FA lie on a line.
Proof. We consider only one case of the relative positions of the points on
the circle (or a conic). The other cases are treated similarly.
Using a projective transformation we can transform the conic into a
circle. We then have the following configuration (Figure 2.24).
The points A, B, C, D, E and F lie on a circle. Suppose the lines AB
and DE intersect at X, the lines BC and EF at Y, and AF and CD at Z.
We want to show that X, Y and Z lie on a straight line.
46 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.24
The angles BAF and BCF are equal since they subtend the same arc.
Similarly, the angles CDE and CFE are equal. Moreover, the triangles
AZD and CZF are similar. Consider the similarity transforming triangle
AZD into triangle CZF. Then X will transform into X', which is isogonally
conjugate to Y with respect to the triangle CZF (in view of the equality of
the above angles). Therefore LAZX = LCZX' = ZFZY, but this means
that X, Z and Y lie on a line.
FIGURE 2.25
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 47
FIGURE 2.26
1c. The sum of the squared distances from a point P to the vertices of
a triangle attains its minimum when P is the intersection of the medians.
The sum of the squared distances from P to the sides attains its minimum
at the Lemoine point.
It is not difficult to see that this statement can be deduced from the
previous one.
2. Brocard points. It turns out that inside any triangle ABC there
is a point Br1 such that LBABrl = LCBBrl = LACBrl. If Br2 is the
isogonal conjugate of Brl, then obviously LABBr2 = LBCBr2 = LCABr2
(Figure 2.27).
FIGURE 2.27
These two points are called the first and the second Brocard points,
respectively. The ellipse tangent to the sides of the triangle and having foci
at these points is called the Brocard ellipse. For an equilateral triangle these
two points coincide with the center of the triangle and the ellipse coincides
with the incircle.
We now prove the existence of Brl. Using the sides of the triangle,
construct the triangles BCA1, B1CA, BC1A similar to it as shown in Fig-
ure 2.28.
Then the circumcircles of these triangles (call them Wa, Wb and wc) in-
tersect in a point. Indeed, let Br1 be the intersection of the circles Wa and
wb different from C. Then
LABr1B = 360° - (LABr1C + LBBr1C)
= 360° - (180° - LAB1C) - (180° - LBA1C)
= LBA1C + LAB1C = LC + LA
= 180° - LB = 180° - LAC1B.
Therefore Br1 lies on wc. We also have
LCBBrl = 180° - LBBr1C - LBCBr1 = LBA1C - LBCBrI
= LBCA - LBCBrI = LACBrl.
Similarly, LCBBrl = LBABrl.
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 49
FIGURE 2.28
That such a point is unique can be easily seen from the construction.
Indeed, suppose that there is another such point X. Then a similar argument
shows that LBXA = 180° - LB, and therefore X lies on the circle w,.
Similarly, it has to lie on the circles w,, and wb, and therefore it coincides
with Brl.
The angle BABr1 is called the Brocard angle of the triangle ABC.
Notice also that the points A, Brl, Al lie on a line. Indeed,
LABr1A1 = LABr1C + LCBr1A1 = 180° - LAB1C + LCBA1
= 180° - LA + LA = 180°.
The Brocard points have some interesting properties.
2a. The pedal and the circumcevian triangles of the Brocard points are
similar to the triangle ABC.
Let A', B' and C' be the projections of Br1 to the sides (Figure 2.29).
The quadrilateral B'AC'Br1 is inscribed, hence LC'B'Br1 = LC'ABr1 =
LACBr1 (the last equality holds because Br1 is the Brocard point). Simi-
larly, LA'B'Br1 = LA'CBrl. Hence LA'B'C' = LA'B'Brl + LC'B'Brl =
LA'CBr1+LACBr1 = LC. A similar argument shows that LB'A'C' = LB
and LA'C'B' = LA. Since the pedal triangle is similar to the circumcevian
triangle, the circumcevian triangle of Brl is similar to the triangle ABC.
2b. Let 0 be the center of the circumcircle of the triangle ABC. Then
OBr1 = OBr2 and LBr1OBr2 equals twice the Brocard angle.
50 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.29
11
FIGURE 2.30
Notice that under the rotation about 0 through the double Brocard
angle, A transforms into C". The same is true for B and C. Therefore, under
the rotation about 0, the triangle ABC will transform into the triangle
C"A"B". Then Brl will be the second Brocard point of the triangle A"B"C"
(LC"A"Brl = LC"A"A = LC"CA = LBr1CA).
Therefore, under the rotation about 0 through the double Brocard an-
gle, Br2 will transform into Brl.
2c. The Lemoine point lies on the circumcircle of the triangle ABr1Br2
and is antipodal to 0.
Consider a projective transformation putting the point Brl at the center
of the circle w circumscribed about the triangle ABC. The triangles ABC
and A"B"C" will clearly transform into the triangles symmetric with respect
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 51
to the center of w, and therefore their Lemoine points will also be symmetric.
On the other hand, these points will be the images of the Lemoine points of
the triangles ABC and A"B"C", hence L, Br1 and L" lie on a line, where
L" is the Lemoine point of A"B"C". Moreover, since the triangles ABC
and A"B"C" can be obtained from each other by rotation through twice the
Brocard angle about 0, the points L and L" are equidistant from the ends
of the chord passing through them. This, together with the preservation
of the cross-ratio under projective transformations, implies that Brl is the
midpoint of the segment LL" and that the angle OBr1L is right. Therefore
Br1 and Br2 lie on the circle with diameter OL and are symmetric with
respect to the line OL.
2d. The Brocard ellipse is tangent to the sides of the triangle formed by
the feet of the symmedians.
Suppose triangles BCA1 and C2AB are similar to ABC and are posi-
tioned as in Figure 2.31.
Al
FIGURE 2.31
Let wi and w2 be their circumcircles. Since Br1 and Br2 lie on wQ and
wc2, respectively, and the angles subtending the arcs CBr1 and ABr2 are
equal, the ratio of the segments CBr1 and ABr2 equals the ratio of the
radii of wa and wc. On the other hand, the ratio of those radii equals the
similarity coefficient of the triangles BCA1 and C2AB, and therefore equals
BC _ BC AB _ BC 2 _ CLb
AC2 AB AC2 - CAB) ALb'
CLb
i.e., CiBr1
ABr2 = ALb
And since LBr1CA = LBr2AC (the Brocard angles),
the triangles CLbBr1 and ALbBr2 are similar. Therefore LBr1LbC =
LBr2LbA. Hence the ellipse with foci at the Brocard points and sum
52 2. SOME RESULTS FROM CLASSICAL GEOMETRY
FIGURE 2.32
The ellipse with foci at those points and tangent to the sides of the
triangle ABC is called the inscribed Steiner ellipse (Figure 2.32).
We prove that its center coincides with the centroid of the triangle. The
centroid corresponds to 3 (za + zb + z,), whereas the centroid of P and Q
corresponds to 2(p+q). We have P'(z) = 3z2-2(za+zb+zc)z+zazbzc, and
therefore, by Viete's theorem, the sum of the roots of the derivative equals
3 (za + zb + zc), i.e., their centroid is (Za + zb + zc), which is what was to
be proved. 3
Remarks. 1. Actually any polynomial of degree greater than one has
the above property: the centroid of the roots of the polynomial and the cen-
troid of the roots of its derivative coincide. The easiest way to prove this is to
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 53
move the centroid to 0. Then the second coefficient of the polynomial, and
therefore the second coefficient of its derivative, will vanish; hence the sum
of the roots of the derivative (and therefore the centroid) will be zero, i.e.,
the centroid will coincide with the centroid of the roots of the polynomial.
2. Consider an affine transformation transforming an equilateral triangle
into the triangle ABC. Then the incircle of the triangle will become an
ellipse and the center of this ellipse will be the centroid of the triangle
ABC. As we will show in Chapter 4, there is a unique conic tangent to the
three given lines, with center at a given point. Therefore this will be the
Steiner ellipse.
Since in an equilateral triangle the tangency points of the incircle are
the midpoints of the sides, the Steiner ellipse will also be tangent to the
sides of ABC at their midpoints.
Similar to the inscribed Steiner ellipse is the circumscribed Steiner el-
lipse; this is the ellipse that passes through the vertices of the triangle and
has its center at the intersection point of the medians. It is the image of the
circumcircle of an equilateral triangle under the affine transformation that
sends it to the triangle ABC.
4. The points of Apollonius and Torricelli. Given a triangle ABC,
the locus of points P such that Ps = BC is the Apollonius circle of the
points A and B that contains the feet of the internal and external bisectors
of the angle C as antipodal points. For points P1, P2, the intersections of
this circle with the similar circle constructed using another pair of vertices,
we have PA BC = PB AC = PZC AB, hence the points Pi lie also on
the third such circle. They are called the Apollonius points of the triangle
ABC. Henceforth we shall denote these points Ap1 (the first Apollonius
point is usually taken to be the one inside the circumcircle) and Ape. Ex-
amining the angles we see that the Apollonius circles are orthogonal to the
circumcircle of ABC. Hence the inversion with respect to the circumcircle
preserves those circles and transforms the Apollonius points into each other.
In particular, the line Ap1Ap2 passes through the center 0 of the circum-
circle. Furthermore, the center of the Apollonius circle passing through C
is the intersection of the line AB with the tangent to the circumcircle at C.
Under the polar correspondence with respect to the circumcircle, these lines
correspond to the intersection point of the tangents to it at A and B and
the point C. Therefore the polars of the centers of the Apollonius circles
are the symmedians, and the pole of the line containing those centers is the
Lemoine point L. Thus L also lies on the line Ap1Ap2 and is the inverse of
the midpoint of the segment AplAp2 with respect to the circumcircle.
We now mention an important property of the Apollonius points which
can be taken as their definition.
The pedal triangles of the Apollonius points are equilateral.
Proof. Let A', B' be the projections of Api to BC and CA. Since the
quadrilateral CA'ApiB' is inscribed in the circle with diameter ApiC, we
54 2. SOME RESULTS FROM CLASSICAL GEOMETRY
have A'B' = ApiC sin C = ApiC AB/2R. This implies at once that all the
sides of the pedal triangle are equal.
It is now easy to see which points are isogonally conjugate to the Apol-
lonius points. Indeed, the perpendiculars dropped from A and B to the
corresponding sides of the pedal triangle intersect at a point Ti that is the
isogonal conjugate of Api. Since the pedal triangle is equilateral, the angle
ATiB equals either 60° or 120°. Thus, at Ti, all the sides of the triangle
are seen at angles 60° or 120°. The points with this property are called the
Torricelli points of the triangle ABC. They can be constructed as follows:
let A', B' and C' be the vertices of the equilateral triangles based on the
sides of the triangle ABC and pointing outside (inside). Then the lines AA',
BY and CC' intersect at the first (second) Torricelli point (Figure 2.33).
FIGURE 2.33
If the angles of the triangle are less than 120°, then the first Torricelli
point lies inside the triangle and the sum of the distances from that point
to the vertices is less than from any other point X in the plane. This can
quickly be checked by turning the triangles AXC and AT1C through 60°
about A (Figure 2.34).
We now mention without proof three more properties of the Torricelli
and Apollonius points.
1) The lines Ap1T1 and Ap2T2 are parallel to the Euler line OH;
2) The line T1T2 passes through the Lemoine point, which implies, as we
will show in 3.3, that the lines Ap1T2 and Ap2T1 intersect at the centroid of
the triangle ABC;
3) LApBrL = 60° (the indexing plays no role since Apl, Ape and L lie
on a line with respect to which Brl and Br2 are symmetric); in particular
2.3. SOME FACTS FROM THE GEOMETRY OF THE TRIANGLE 55
FIGURE 2.34
this means that the triangle OBr1Br2 has the same Apollonius points as the
original triangle.
5. The Gergonne and Nagel points and the homothety centers
of the circumcircle and the incircle.
Definition. Suppose the incircle of a triangle ABC is tangent to the sides
at points Ga, Gb and G,. Then the lines AGa, BGb and CG, intersect in a
single point G (this can be shown with the aid of Ceva's theorem or by using
a projective transformation preserving the incircle and transforming the
intersection of the lines AGa and BGb into the center), called the Gergonne
point.
Draw the reflections of AGa and BGb in the bisectors Al and BI and
find their intersection points Al, B1 with the incircle on the other side of A
and B (Figure 2.35). We have
FIGURE 2.35
FIGURE 2.36
either the distances from X to the other intersections with those circles are
distinct (and therefore the absolute values of the powers are not equal) or
the signs of the powers of points are distinct.
Now we prove the theorem in the case where the circles have no com-
mon points. To that end, we apply a useful trick. Consider two spheres
intersecting our plane along the two circles and having common points. It
is easy to see that such spheres exist. Indeed, take a point not in the plane
and the spheres passing through that point and the two circles. For spheres,
the power of a point can also be defined, and Lemma 2.7 is still true. A
similar argument shows that the set of points whose powers with respect to
the two spheres are equal is the plane passing through the intersection circle
of the spheres (see Figure 2.37). That plane intersects our plane along a line
which is obviously the radical axis of our circles. 0
FIGURE 2.37
Suppose now that three circles are given. If their centers are not on a
line, then there are two pairs of those circles whose radical axes intersect.
The powers of their intersection point with respect to all three circles are
equal and therefore the third radical axis also passes through that point,
called the radical center of the three circles. If the centers of the circles are
on a line, then the radical axes are either parallel or coincide. In the latter
case the circles are said to be coaxial.
The set of all circles coaxial with two given circles is called a pencil.
If the circles defining the pencil intersect at two points, then the pencil
consists of all circles passing through these points. Such a pencil is said to
be hyperbolic. If the two circles are tangent, then any circle of the pencil is
tangent to their common tangent line at the same point. Such a pencil is
said to be parabolic. Finally, two nonintersecting circles give rise to a pencil
of the type shown in Figure 2.38. Such a pencil is said to be elliptic. Notice
2.4. RADICAL AXES AND PENCILS OF CIRCLES 59
that two of the circles of an elliptic pencil degenerate into points, called the
limit points of the pencil.
FIGURE 2.38
For any point on the radical axis which is outside the circles, the tangents
to the circles of the pencil passing through that point are equal. Hence
the circle centered at that point with radius the length of the tangent is
perpendicular to all the circles of the pencil. All such circles form another
pencil (Figure 2.39), and any two of them uniquely determine the original
pencil. It now follows that the inversion in an arbitrary circle transforms
pencils into pencils; moreover any pencil containing the circle of inversion
transforms into itself. In particular, the limit points of an elliptic pencil
transform into each other under inversion in any circle of that pencil. Notice
also that the inversion centered at a limit point transforms the circles of the
perpendicular pencil into lines. Therefore the original pencil is transformed
into a pencil of concentric circles.
FIGURE 2.39
FIGURE 2.40
Using this theorem one can easily prove the theorem of Poncelet (for
a pencil of circles) without algebraic arguments. In its general from, the
theorem of Poncelet will be proved in 3.3.
FIGURE 2.41
Proof. It suffices to show that AiBi is tangent to some fixed circle from the
pencil. Indeed, if A0 coincides with An, then the tangent to the circle passing
through A0 coincides with the tangent passing through An (assuming the
tangents run in the appropriate direction). Therefore their intersections
with wo must coincide, but these are Bo and Bn.
Suppose the lines A0A1 and BOB1 are tangent to wi at X and Y, respec-
tively (Figure 2.42). Let Z be the intersection of these lines. The triangle
XZY is isosceles (because ZX and ZY are simply the tangents to wl pass-
ing through Z). Therefore the angles X and Y in thattriangle are equal.
62 2. SOME RESULTS FROM CLASSICAL GEOMETRY
Moreover the angles B1A1AO and BIBOAO are also equal. Thus the trian-
gles XQA1 and YPBO are similar, where P and Q are the intersections of
the line XY with the segments AOBO and A1B1. Hence the angles PQA1
and QPBO are equal, and therefore there is a circle w' tangent to AoBo and
A1B1 at P and Q, respectively.
FIGURE 2.42
FIGURE 2.43
Besides the Poncelet theorem, the properties of radical axes also allow
for a proof of Brianchon's theorem in its general form. First, we need to
state it.
FIGURE 2.44
axis of wl and w2, and CF is the radical axis of w2 and w3. Therefore AD,
BE and CF intersect at a single point, namely, the radical center of wl, w2
and w3.
Chapter 3
Projective Properties of
Conics
3.1. The cross-ratio of four points on a curve. Parametriza-
tion. The converses of Pascal's and Brianchon's theorems
Projective equivalence of conics means that all the properties of the circle
mentioned in the Introduction are also true for conics. In particular, for four
points A, B, C, D on a conic, the cross-ratio of the lines XA, XB, XC,
XD does not depend on the choice of a point X on the conic. This ratio
is called the cross-ratio of A, B, C, D. Clearly, the cross-ratio is preserved
under projective transformations.
We now choose and fix some point P of the conic and some line 1 that
does not pass through P. Given a point X on the conic, consider the inter-
section point X' of PX and l (for the point P itself, we take the intersection
of l with the tangent at P). Clearly, this correspondence is one-to-one and
preserves the cross-ratio. Now any standard correspondence between the
points of 1 and the real numbers gives rise to a parametrization of the conic.
It is not difficult to see that under such a parametrization the coordinates
of X are rational functions of the parameter.
Theorem 3.1 (The converse of Pascal's theorem). For any six points
Xi, i = 1, ... , 6, such that the intersections of the lines X1X2 and X4X5,
X2X3 and X5X6, X3X4 and X6X1 are on a straight line, there is a conic
passing through all the Xi.
Proof. We use the fact that for any five points in general position, there is
a unique conic containing them. Let a be such a conic for Xi, i = 1, ... , 5.
Let A, B, C be the intersections of the lines X1X2 and X4X5, X2X3 and
X5X6, X3X4 and X6X1, and let Y be the intersection point of a and BX5
different from X5. By Pascal's theorem, the intersection of X3X4 and X1Y
lies on AB, i.e., coincides with C. Hence Y coincides with X6. 0
65
66 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.1
FIGURE 3.2
FIGURE 3.3
68 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.4
Indeed, applying this construction to the point A' and to the circle we
have the line a'. Since projective transformations preserve the intersections
and tangencies of lines and conics, the line a does not depend on the chosen
projective transformation. There is another way to construct a: this is the
line connecting the intersections of X1Y1 with X2Y2 and of X1Y2 with X2Y1
(Figure 3.5). In particular, if A is the center of an ellipse or a hyperbola,
we get the line at infinity. Notice that the last construction also applies
to degenerate curves of degree two; moreover, the constructed line passes
through the common point 0 of the lines ll and 12 comprising the curve,
and the cross-ratio (1112; OAa) of the lines equals 1 (Figure 3.6).
FIGURE 3.5
The defined correspondence between points and lines is called the polar
correspondence with respect to a given conic; a is called the polar of A and
A is the pole of a. Clearly, all the properties of the polar correspondence
mentioned before are still true.
3.2. POLAR CORRESPONDENCE. THE DUALITY PRINCIPLE 69
FIGURE 3.6
FIGURE 3.7
In particular, if the conic is an ellipse or a hyperbola, then the midpoints
of all chords parallel to a fixed line lie on a line passing through the center of
the conic (the directions of this line and the fixed line are said to be conjugate
with respect to the conic); if the conic is a parabola, then the midpoints lie
on a line parallel to its axis (see Problem 10). The dual assertion is also
true: if tangents a and b to a conic and two arbitrary lines p and q meet at
one point, then the pole of p lies on q if and only if (ab;pq) = I.
The duality principle also remains true. Hence, for example, the converse
of Brianchon's theorem is a consequence of the converse of Pascal's theorem.
Furthermore, for any five lines in general position, there is a unique conic
tangent to them.
Definition. The dual curve of a smooth curve is the set of duals to all the
tangents of this curve.
An example of a curve and its dual curve is shown in Figure 3.8.
The following theorem provides an important property of the duality
operation.
Theorem 3.3. Let R(y) be the dual curve of a curve -y. Then R(R(y)) = y.
Proof. Suppose a point X moves along y toward A. Then, clearly, the
intersections of the tangents at X and A (call them x and a, respectively)
70 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.8
FIGURE 3.9
0
01
FIGURE 3.10
Thus we need to show that the set of all such lines is tangent to a conic.
Consider the reflection 0' of 0 in 02, and the reflection 0" of 0 in X. It
is easy to see that the length of the segment O'O" equals the diameter of
W2. Suppose O'O" intersects p(X) at a point Y. Then, since p(X) is the
midpoint perpendicular to 00", the segments YO and YO" are of equal
length. Moreover, the angles formed by the line p(X) with YO and YO'
are equal. Therefore p(X) is tangent at Y to the ellipse with foci 0 and 0'
and major semi-axis equal to the diameter of w2. Moreover, it is easy to see
that, as X moves along w2, the point Y sweeps the entire ellipse.
Thus we have constructed a conic which is the polar curve of our circle.
If 0 is outside wl, then identical arguments show that the polar curve
is a hyperbola, and in the case when 0 lies on wl, the polar curve is a
parabola.
Because of the projective equivalence of conics, the theorem just proved
can be generalized.
Theorem 3.5. The polar of a conic with respect to another conic is also a
conic.
The only thing that is not clear is why any two conics can be made into
circles by a projective transformation. In general, this is not true (although
this is true over complex numbers). But this is possible if the conics intersect
at no more than two points. This can easily be achieved by scaling (with
center at the center of the conic) the conic with respect to which we perform
a polar transformation so that it would intersect our conic at most at two
points. The dual curve will then scale in the opposite way. Now any two
72 3. PROJECTIVE PROPERTIES OF CONICS
conics intersecting at most at two points can be made into two circles using
a projective transformation.
In fact, this theorem can be proved by using only the theorems of Pascal
and Brianchon. Choose and fix five points X1, X2, X3, X4 and X5 on the
conic a whose polar curve we consider. Then the polars of those five points
are tangent to some other conic, which we denote al. Suppose a point X is
moving along a. Then the corollary of Pascal's theorem can be applied to
the points Xl,... , X5 and X. Hence the corollary of Brianchon's theorem
applies to the polars of those points. But the converse of Brianchon's theo-
rem implies that all these six lines are tangent to a conic. It could only be
the conic al, since five of the lines (the polars of Xi, i = 1, . . . , 5) can be
tangent to only one conic. Thus the polars of all points on a are tangent
to al. Reversing the argument, one can easily show that the conic al is
traversed completely.
Finally, we mention yet another approach to defining conics and polar
correspondences. Suppose we have a one-to-one correspondence between
points and lines in the projective plane which has the duality property, i.e.,
if a point A belongs to the image of a point B, then B belongs to the image of
A. Then the set of points belonging to their own images is a conic (possibly,
imaginary) and the polar correspondence with respect to that conic coincides
with the given correspondence.
We now show that a focus and the corresponding directrix of a conic
are polar to each other. In fact, this has already been established for the
parabola. We prove this for the remaining conics.
Theorem 3.6. A focus and the corresponding directrix of a conic are polar
to each other (Figure 3.11).
FIGURE 3.11
Proof. Consider a focus F1 and its polar 1. We prove that for any two points
X and Y on the conic, the ratios of the distances to F1 and 1 are equal. Let
S be the intersection of the lines XY and 1. Let Z denote the intersection
of the tangents to the conic at X and Y. By the properties of the polar
transformation, F1Z is the polar of S. By the corollary of Theorem 1.2,
the angle SF1Z is right. Moreover, Theorem 1.4 implies that F1Z is the
bisector of the angle X F1Y. Therefore Fl S is the bisector of the exterior
3.2. POLAR CORRESPONDENCE. THE DUALITY PRINCIPLE 73
------------ -
t
FIGURE 3.12
The polar plane of the point S with respect to the sphere SZ is the plane
a, and the polar plane of the point F is the plane 7r. Hence the polar of
the line SF is the line 1. It is clear that the pole of 1 with respect to the
intersection circle of S2 and v is the intersection of the line SF and the plane
Q. Therefore F is the pole of 1 with respect to the intersection ellipse of our
cone and the plane 7r (consider the projection of a to it from the point S).
An interesting example of assertions interchanged by a polar transfor-
mation is Theorem 1.11 and Problem 19.
Indeed, let H be the orthocenter of a triangle ABC. The polar transfor-
mation with respect to a circle w with center H maps the triangle ABC into
a homothetic (with respect to H) triangle A'B'C'. If a parabola is tangent
to the sides of the triangle ABC, then the conic dual to it with respect to w
will pass through A', B' and C' and also through H, because it is the pole
of the line at infinity. By Problem 19, this conic is an equilateral hyperbola.
The intersection points of this hyperbola and the line at infinity give rise to
74 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.13
Similar arguments allow one to deduce the assertion of Problem 19 from
Theorem 1.11. Therefore these assertions are dual to each other.
Now we mention a few more results related to the Ceva triangles of
triangles inscribed in a conic.
Theorem 3.7. A triangle ABC is self-polar (i.e., its sides are the polars
of the corresponding vertices) with respect to a conic if and only if it is the
Ceva triangle of a point on the conic with respect to a triangle inscribed in
the conic.
FIGURE 3.14
become a conic centered at A (since the center of the conic is the pole of
the line at infinity). Consider the rectangle inscribed in the conic with sides
parallel to the directions given by B and C (such a rectangle exists because
the triangle ABC is self-polar). Its vertices can be viewed as a triangle and a
point on the conic for which the triangle ABC is self-polar (Figure 3.14). 0
Theorem 3.8. Suppose we are given a triangle ABC and a point Z. For
an arbitrary line passing through Z, let A' and B' be its intersection points
with BC and AC. Then the locus of the intersections of the lines AA' and
BY is a conic passing through A, B and C and tangent to the lines AZ
and BZ.
Proof. Apply a projective transformation making the triangle ABC into a
right isosceles triangle (AC = BC) and sending the point Z to infinity in
the direction perpendicular to AB. Then the triangles AA'P and B'BP,
where P is the intersection of AB and A'B', are equal, and therefore the
lines AA' and BB' are perpendicular; i.e., their point of intersection lies on
the circumcircle of the triangle ABC. Moreover, the lines BZ and AZ are
tangent to that circle at A and B. 0
Now let Y be the intersection of the tangents to the conic at A and C.
Consider the intersections A' and C' of the lines passing through Y with
BC and AB. Then the intersection of the lines AA' and CC' lies on the
conic. This means that there is family of triangles A'B'C', with vertices on
the corresponding sides of the triangle ABC, with the following properties.
1. For each triangle in the family, the lines AA', BB' and CC' intersect
at a single point. The set of all such points is a conic passing through A, B,
C.
2. All lines A'B' pass through the pole of the line AB with respect to
the circumscribed conic. Similarly, all lines A'C' pass through the pole of
the line AC, and all lines B'C' pass through the pole of the line BC.
Projective properties of conics may be useful for proving results seem-
ingly unrelated to conics. As an example, we have the following.
Theorem 3.9. Suppose we are given a triangle ABC and points P and Q,
and suppose that the lines AP, BP and CP intersect the respective sides of
the triangle at points A1, B1, C1 and that the lines AQ, BQ, CQ intersect
the respective sides at points A2, B2, C2. Let C3, C4 be the intersections of
the lines CC1 and A2B2, CC2 and A1B1, respectively; the points A3, A4,
B3, B4 are defined similarly. Then the lines A1A4, A2A3, B1B4, B2B3,
C1C4i C2C3 intersect at a single point (Figure 3.15).
FIGURE 3.15
dropped from P and Q lie on the circle. Examining the quadrilateral formed
by those feet and the points P and Q, we see that the orthocenter of the
triangle PQC1 lies on the polar of C1, i.e., on A1B1. Since it also lies on
the line QC, it coincides with C4. Thus the line C1C4 is perpendicular to
the diameter PQ, i.e., it passes through the pole of the diameter. Similarly,
the remaining five lines also pass through the pole of PQ, hence the pole is
the point mentioned in the theorem. 0
Using duality we can prove the following nice theorem.
Theorem 3.10 (Fregier). Suppose we are given a conic and a point P on
it. Then all the chords seen from P at a right angle pass through a single
point (Figure 3.16).
FIGURE 3.16
Proof. Apply the polar correspondence with respect to a circle with cen-
ter P. Since the given conic passes through P, its transform will be a
parabola. The perpendicular lines passing through P will transform into
the two points at infinity corresponding to the perpendicular directions, and
their second intersections with the conic will become perpendicular tangents
3.2. POLAR CORRESPONDENCE. THE DUALITY PRINCIPLE 77
FIGURE 3.17
to the parabola. Since the intersection of those tangents lies on the directrix,
the corresponding chord passes through its pole P'.
Clearly, P is the intersection of the diameter symmetric to the diameter
passing through P and the normal to the conic at P (Figure 3.17).
It is not difficult to see that P' cuts the diameter in the ratio equal to
that of the squares of the axes of the conic, and therefore when P moves
along the conic, P' sweeps a conic which is homothetic to the original conic
with respect to its center (if the original conic is a parabola, then the new
conic is obtained from it by parallel translation (Figure 3.18)).
FIGURE 3.18
FIGURE 3.19
78 3. PROJECTIVE PROPERTIES OF CONICS
Problem 21. Let C be the center of a conic which is the polar of a circle
a with respect to a circle w. Prove that the polar of C with respect to w
coincides with the polar of the center of w with respect to a.
Problem 22. 1. Prove that the directions conjugate with respect to an
equilateral hyperbola are symmetric with respect to its asymptotes.
2. Prove that the angle between concentric equilateral hyperbolas is
equal to twice the angle between their asymptotes.
Problem 23. What kind of curve is enveloped by the sides of the rhombi
inscribed in a fixed ellipse?
hyperbola. It now follows from the theorem on pencils of conics that if two
equilateral hyperbolas intersect at four points, then any conic in the pencil
determined by those points will also be an equilateral hyperbola. Thus we
have yet another proof of the fact that a conic circumscribed about a triangle
is an equilateral hyperbola if and only if it passes through the orthocenter.
FIGURE 3.20
The theorem on pencils of conics allows one to define the pencil as the set
of conics passing through four given points A, B, C, D in general position.
Moreover, for any point X different from A, B, C, D, there is exactly one
conic of the pencil passing through X.
Some of the points defining the pencil can be imaginary. For example,
any circle intersects the line at infinity in two fixed complex points so that
a hyperbolic pencil of circles is defined by those points and two common
80 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.21
If all points defining the pencil are distinct, then there are three degen-
erate curves in that pencil: AB U CD, AC U BD and AD U BC.
Now we describe the various types of pencils in more detail.
1. Pencils passing through four distinct points (Figure 3.20). Elliptic
and hyperbolic pencils of circles belong to this type.
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 81
2. Pencils passing through four points two of which coincide, i.e., pencils
tangent to a given line at a fixed point (Figure 3.21). Parabolic pencils of
circles belong to this type.
3. A pencil defined by two coinciding pairs of points. It consists of
conics tangent to two given lines at two given points (Figure 3.22). Pencils
of concentric circles or parabolas with equation y = ax2 belong to this type.
FIGURE 3.22
FIGURE 3.23
FIGURE 3.24
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 83
In addition to pencils defined by four points one can consider dual pen-
cils, i.e., sets of conics tangent to four given lines. Dual pencils are then
classified according to the number of coinciding defining lines. If two lines
coincide, then all the conics of the pencil are tangent to them, and therefore
to each other, at a point.
FIGURE 3.25
84 3. PROJECTIVE PROPERTIES OF CONICS
If three of the lines coincide, then the conics of the pencil osculate. If
all four lines coincide, then the conics hyperosculate. Notice that double
tangent and hyperosculating pencils are self-dual, i.e., are transformed into
themselves under the polar correspondence with respect to any conic in the
pencil. The duality principle allows one to state, for each assertion concern-
ing usual pencils, the corresponding assertion concerning dual pencils, and
vice versa.
FIGURE 3.26
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 85
FIGURE 3.27
Using the theorem on pencils of conics one can prove the following nice
result.
Theorem 3.11 (The four conics theorem). Suppose we are given three
conics al, a2i a3, and let Pi, Ql, Pi, Qi be the intersections of a2 and
a3i P2, Q2, P2, Q2 the intersections of al and a3; and P3, Q3i P3, Q3 the
intersections of a2 and al. If the points P1, Qi, P2, Q2, P3, Q3 lie on one
of the conics, then the lines P11Q1', P2Q2, P3Q3 intersect at a single point
(Figure 3.28).
FIGURE 3.28
Theorem 3.12 (The three conics theorem). Suppose three conics have
two common points. Then their common chords passing through the remain-
ing intersections of each pair meet at a single point (Figure 3.29).
FIGURE 3.29
To prove this claim it suffices to transform the common points of the
conics into the intersection points of the line at infinity and circles. Then all
three conics will transform into circles and the desired assertion will follow
from the existence of the radical center.
Now let ao be the conic passing through the points Pi, Qi, and let
Fi(x, y) = 0 be an equation of ai. Since the degenerate conic consisting of the
lines P2P3 and Q2Q3 belongs to the same pencil as ao and al, we may assume
that its equation is of the form F0 = Fl. Similarly, the conics consisting of
the lines P1Q1, P3Q3 and P1Q1, P2Q2, will be given, respectively, by the
equations F0 = F2 and F0 = F3. Therefore for each point on the line P3Q3
we have F1 = F2. Since this also holds for the points P3 and Q3, this is
an equation of the degenerate conic consisting of the lines P3Q3 and P3Q3'
Accordingly, the conics consisting of the lines P2Q2 and P2Q2, is given by
the equation F1 = F3, and the conic consisting of the lines P1Q1 and P1Q' is
given by the equation F2 = F3. Thus these three conics belong to the same
pencil. Three of the points determining this pencil can be found using the
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 87
three conics theorem applied to the triples ao, al, a2; ao, al, a3; and ao,
a2, a3: they are the intersections of the triples of lines P1Q1, P2Q2, P3Q3;
P1Q1, P3Q3, PPQ'; and P2Q2, P3Q3, PiQi. Hence the fourth point belongs
to all the lines P'Q. Similar arguments prove the converse: if each of the
four triples of lines P1Q1, P2Q2 and P3Q3; P1Q1, P3Q3 and P2Q2; P2Q2,
P3Q3 and PiQi; and PiQi, P2Q2 and P3Q3 intersects at a single point, then
the points P1, Q1, P2, Q2, P3, Q3 (as well as P1i Q1i P2, Q2, P3, Q'3 and
the two similar sextuples) lie on a conic. O
Using the duality principle we have the following results.
Theorem 3.13 (Dual to the three conics theorem). Suppose three
conics are tangent to two given lines. Then the intersections of the common
tangents to each pair of conics which are different from the given lines lie
on a straight line (Figure 3.30).
FIGURE 3.30
Theorem 3.14 (Dual to the four conics theorem). If two of the com-
mon tangents to each pair of three given conics are tangent to the same
conic, then the intersections of the other two tangents to each pair lie on a
straight line (Figure 3.31).
We now mention some important properties of pencils.
Theorem 3.15. Let A, B, C, D be four distinct points, and X, Y, Z the
intersections of the lines AB and CD, AC and BD, AD and BC. Let P
be a point different from X, Y and Z. Then the polars of P with respect to
all the conics of the pencil determined by the points A, B, C and D pass
through a single point.
Here is an interesting special case. If the points A, B, C and D form
an orthocentric quadruple (i.e., each point is the orthocenter of the trian-
gle formed by the remaining points), then the obtained point is isogonally
conjugate to P with respect to the triangle XYZ.
Indeed, the polar of P with respect to the degenerate curve formed by
the lines AB and CD is the line symmetric to XP with respect to AB. Since
AB and CD are the bisectors of the angle YXZ, this line passes through
88 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.31
FIGURE 3.32
set of points satisfying the above condition is a straight line. Clearly, the
midpoints of the diagonals belong to that line.
Suppose now that a conic with foci F1 and F2 is inscribed in the quadri-
lateral ABCD. Since its center is the midpoint of the segment F1F2i the
assertion of the theorem is equivalent to the assertion that SF1 AB + SF1 CD +
SF2AB + SF2CD = SF1BC + SF1DA + SF2BC + SF2DA
Let F1' be the reflection of F1 in AB. Then
But the points F1 and F2 are isogonally conjugate with respect to ABCD,
and therefore LF11AF2 = LF1AB + LF2AB = LA, LFFBF2 = LB and
FIGURE 3.33
90 3. PROJECTIVE PROPERTIES OF CONICS
Q\ !
FIGURE 3.34
This, together with similar equalities, implies that the left- and the right-
hand sides of the desired equality are equal to
Proof. Consider the pencil of conics giving rise to the conjugation in ques-
tion. In that pencil, choose a conic with respect to which the polar of X
coincides with the line X'Y'. Then the polar of Y' passes through X, i.e., it
coincides with XY, and the pole of the line XY' is the intersection point of
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 91
the lines XY and X'Y'. Therefore the conjugate of that point lies on XY'.
Similarly, it lies on X'Y.
Theorem 3.17. The poles of a fixed line with respect to all the conics of
the pencil defined by points A, B, C and D form a conic.
Proof. Transform the given line into the line at infinity. Then its poles will
be the centers of the conics of the pencil. It follows from the converse to
Pascal's theorem that the midpoints K, L, M and N of the sides of the
quadrilateral ABCD belong to the set of the centers. Hence it suffices to
show that for the center 0 of any conic of the pencil, the cross-ratio of the
lines OK, OL, OM and ON is the same. This cross-ratio equals the ratio
of the poles of those lines, which are the intersection points of the line at
infinity with the sides of the quadrilateral ABCD and are independent of
the choice of the conic.
Notice that the center of a degenerate curve of degree two is the inter-
section of the lines making that curve. Therefore the conic mentioned in
Theorem 3.17 always passes through the intersections of the lines AB and
CD, AC and BD, AD and BC.
Similar to Theorem 3.15, Theorem 3.17 also has an important special
case.
Corollary. Suppose we are given a triangle ABC and a line 1 not passing
through its vertices. Then the isogonal (isotomic) conjugate of 1 is a conic
passing through A, B and C.
This corollary yields another proof of the assertion of Problem 19. The
isogonal conjugates of the conics passing through the vertices of the triangle
are lines. The infinite points of those conics transform into points on the
circumcircle, and the points in the perpendicular direction transform into
antipodal points (as is easy to check). Hence those lines pass through the
center of the circumcircle of the triangle. The isogonal conjugate of that
center is the orthocenter of the triangle; i.e., the conic must contain the
orthocenter of the triangle. The converse is proved similarly.
Theorem 3.15 has another nice proof under the assumption that the
pencil in question consists of circles.
Notice that the radical axes of P and the circles of the pencil W intersect
at a single point, which we denote Q. This is clear, since on the radical axis
of W, the powers of points with respect to all circles are equal. Therefore
the desired P is a point on the radical axis whose power with respect to
some circle in the pencil equals the square of the distance to P. The polar
of P with respect to any circle is parallel to the radical axis of P and the
circle and is twice as far from P (Figure 3.35). Thus all the polars of P with
respect to the circles of W pass through the reflection of P in Q.
Problem 24. Prove that an equilateral hyperbola is self-dual with respect
to the circle that is tangent to the hyperbola at its vertices.
92 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.35
Problem 27. Two points of a conic lie on two circles one of which intersects
the conic at points X1 and Y1, and the other at points X2 and Y2. Prove
that the lines X1Y1 and X2Y2 are parallel.
Problem 28. Prove that for any quadrilateral, the midpoints of its sides
and of the diagonals, as well as the intersections of the diagonals and of the
opposite sides, lie on a conic. What kind of conic is it if the vertices of the
quadrilateral form an orthocentric quadruple?
Problem 29. Prove that the centers of the conics circumscribed about a
quadrilateral ABCD form an equilateral hyperbola if and only if the quadri-
lateral ABCD is inscribed.
Problem 30. Suppose we are given three circles, each lying outside the
other two. Prove that the common inner tangents to each pair of circles
form a hexagon whose main diagonals intersect at a single point.
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 93
Poncelet's theorem.
Consider now the pencil determined by points A, B, C, D and a line 1 not
passing through these points. If one of the conics of the pencil intersects 1
at P, then it intersects 1 at another point P (which may coincide with P).
The transformation P -* P will be called the involution of 1 defined by the
pencil. Applying Pascal's theorem to the hexagon ABCDPP', we obtain a
method of constructing P', shown in Figure 3.36.
FIGURE 3.36
KL. Moreover, there is a conic of the pencil.F tangent to the lines AD and
BC at their intersections with KL.
2. Suppose that points A, B, C and D lie on a conic ao and that the
conic al is tangent to AB at a point K and to CD at a point M. Then
there is a conic of the pencil F tangent to the lines AC and BD at their
intersection points with KM (Figure 3.37).
FIGURE 3.37
Proof. 1. Let M be the second intersection point of al and KL, and let
g be the pencil containing al and the degenerate curve consisting of the
lines CM and AB. On the line AC, the pencils F and 9 give rise to the
same involution defined by the points A and C and the intersections of AC
with al (which are not necessarily real). Therefore the point L is a double
involution point defined by the pencil 9, i.e., 9 contains the double line KL.
Therefore all conics of 9, including al, are tangent to CM at M. Let D
be the second intersection point of CM and ao. Applying the foregoing
argument to the points B, C and D we have that a2 is tangent to the line
BD at its intersection point with KL. Next we determine the intersection
point of the lines AD and KL and take the conic of F passing through it.
The same argument shows that that conic is tangent to the line AD.
The fact that it is also tangent to BC and assertion 2 of the theorem
are proved similarly.
Corollary. Suppose the line AB is tangent to the conic al at X and the
line AC is tangent to the conic a2 at Y. Then there are exactly two conics
of the pencil .F tangent to BC at points Z1 and Z2; moreover X, Y and Z1
lie on a straight line and the lines AZ2, BY and CX intersect at a single
point.
Now we can prove Poncelet's theorem for n = 3. Suppose the lines
A0A1i A1A2, A2A0 are tangent to the conics a1, a2, a3 at points X1, X2,
X3 that do not lie on a straight line, and the line BOB, is tangent to the
conic al at a point Y1. By Lemma 3.8, there is a conic a' in .F tangent to
the lines A0B0 and A1B1 at their intersection points Z0 and Z1 with X1Y1,
and a point B2 on ao such that a2 is tangent to B1B2 and a' is tangent to
B2A2 at their intersection points Y2 and Z2 with Z1X2. Moreover, there is a
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 95
conic a" tangent to the lines A2A0 and B2B0 at their intersection points K
and Y3 with Z2Z0. Applying Desargues' theorem to the triangles A0A1A2
and Z0Z1Z2, we see that the points X1, X2, K are not on a straight line.
Therefore K = X3 and all = a3.
Now suppose that n is arbitrary. Use the points A0, Al and A2 to
construct a conic Q' tangent to the lines A0A2 and B0B2. Since the sides of
the polygon A0A2 ... An_1 are tangent to the conics a', a3, ... , an, we can
make the induction step from n - 1 to n. 0
Conics with the common focus and directrix.
Consider a family Q of conics with a fixed focus F and the corresponding
directrix 1.
By Theorem 3.15, the polar transformations with respect to those conics
act the same way on the lines passing through F. More precisely, a line a
transforms into the intersection of the perpendicular to a at F with the
line 1.
Using a projective transformation, make one of our conics into a circle
such that F will transform into its center F'. Then the directrix will trans-
form into the polar of the center, i.e., the line at infinity. What happens
with the other conics under such a transformation?
They will transform into the pencil of circles with center F'!
Indeed, they will transform into conics such that the polar of the line at
infinity is F'. But for any conic, the pole of the line at infinity is its center,
and therefore the center F' is the center of all such conics. Moreover, the
pole of any line passing through F' must be the point on the line at infinity
in the direction perpendicular to that line (because this is true for a circle,
and the dual transformation on the lines passing through F' is the same for
all conics). Obviously, this is possible only if all the conics are circles.
Notice that concentric circles transform into each other under the dual
polar transformation with respect to one of those circles. This property is
also preserved under the dual transformation! Thus we have proved the
following result.
Theorem 3.19. The polar transformation with respect to one of the conics
of the family Q (which is a family with fixed focus and directrix) leaves the
family Q unchanged; i.e., the conics of this family transform into conics of
the same family.
The fact that this pencil is projectively equivalent to a pencil of concen-
tric circles allows one to describe the action of this transformation on the
conics.
Theorem 3.20. Let R be the polar transformation with respect to some
conic of Q. Suppose a point X on the conic a (from Q) transforms into
the line R(X) tangent to R(a) at Y. Then the points X, Y and F lie on a
straight line (Figure 3.38).
96 3. PROJECTIVE PROPERTIES OF CONICS
FIGURE 3.38
FX FAY FX F1Z
is to be proved. Note that the left-hand side equals (XY; F1F) (X Z; F1F),
and therefore it does not change under projective transformations. It re-
mains to show that this equality holds when the conics are concentric circles.
Thus we perform a projective transformation making a, al and a2 into
three concentric circles a', ai and a2 with center F' (which is the image of
F). The points X, Y, Z and F1 transform into points X', Y', Z' and F1
lying on a straight line (which also contains F'), and Fi transforms into a
point at infinity.
By the definition of the polar transformation with respect to a circle,
F'Y' F'Z' = F'X'2. Hence
F'Y'F(X' F'Y' oo
F'Z'
F'X'=1.
F'X'
Notice also that the conics mentioned in the generalized Fregier theorem
can be obtained from Q by the dual transformation. Therefore they form a
3.3. PENCILS OF CURVES. PONCELET'S THEOREM 97
pencil of the third kind and, using a projective transformation, we can make
them into concentric circles.
Problem 31. Prove that the foci different from F of two conics from the
pencil Q which are dual with respect to a parabola are symmetric with
respect to 1.
Chapter 4
Euclidean Properties of
Curves of Second Degree
4.1. Special properties of equilateral hyperbolas
Recall that a hyperbola is said to be equilateral if its asymptotes are per-
pendicular. In the previous chapter we gave several proofs that the conic
circumscribed about a triangle is an equilateral hyperbola if and only if it
passes through the orthocenter of the triangle. In this section we establish
some other interesting properties.
Theorem 4.1. The centers of all equilateral hyperbolas passing through the
vertices of a triangle ABC lie on the Euler circle of the triangle.
FIGURE 4.1
FIGURE 4.2
Now consider the triangle ABC and the point P. The circles symmet-
ric to the circumcircles ABP, BCP, CAP with respect to AB, BC, CA,
intersect at a single point. This will be the point P' symmetric to P with
respect to the center of the hyperbola ABCP. Indeed, the previous asser-
tion implies that the circles ABP and ABP' have equal radii; i.e., they are
symmetric with respect to AB.
Theorem 4.2. Suppose we are given a triangle ABC and a point P different
from its orthocenter. Then the centers of the incircle and the excircles of
the Ceva triangle of P with respect to the triangle ABC lie on an equilateral
hyperbola passing through A, B, C and P.
Proof. This property is a special case of the following fact.
4.1. SPECIAL PROPERTIES OF EQUILATERAL HYPERBOLAS 101
Lemma 4.9. Suppose we are given two triangles A1B1C1 and A2B2C2, and
let A', B' and C' be the intersections of B1C1 and B2C2, C1A1 and C2A2,
A1B1 and A2B2, respectively. If the triangle A'B'C' is in perspective with
the triangle A1B1C1 as well as with A2B2C2 (from the centers of perspective
D1 and D2), then the points A1, B1, C1, D1, A2, B2, C2, D2 lie on a conic
(Figure 4.3).
FIGURE 4.3
Proof. By Theorem 4.2, the centers Ia, Ib, I,' of the excircles of the Ceva
triangle lie on the hyperbola. Since the Ceva circle is the nine-point circle
of the triangle I ,IbI,, it passes through the center of the hyperbola.
102 4. EUCLIDEAN PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 4.4
FIGURE 4.5
To prove that the circumcircle of the triangle A'B'C' also passes through
the center Q of the hyperbola, it suffices to show that the angles A'C'B' and
A'QB' are equal.
Notice that LDC'A' = ZDBA' since the quadrilateral C'BA'D is in-
scribed. The segment B1C1 is a midline of the triangle DBC and therefore
LDB1C1 = LDBA'. Since the points D and A' are symmetric with respect
to B1C1, the angles DB1C1 and A'B1C1 are equal. The point Q, as the
center of an equilateral hyperbola, lies on the Euler circle of the triangle
BCD. Therefore LA'B1C1 = LA'QC1. Thus
LDC'A' = LDBA' = LDB1C1 = LA'B1C1 = LA'QC1.
4.1. SPECIAL PROPERTIES OF EQUILATERAL HYPERBOLAS 103
FIGURE 4.6
Proof. Parts 1 and 2 follow from Theorem 3.9. We now prove part 3. Since
the set of the centers of perspective of the triangles from the Feuerbach
family lies on an equilateral hyperbola (it is called the Feuerbach hyperbola),
their Ceva circles pass through the center of the hyperbola. Their pedal
circles also pass through that center. But the pedal circles of the points I
and H have a single common point-the Feuerbach point. Therefore it is
the center of the hyperbola.
104 4. EUCLIDEAN PROPERTIES OF CURVES OF SECOND DEGREE
ABT1, BCT, and CAT, pass through T2. Therefore the midpoint of the
segment T1T2 is the center of the Kiepert hyperbola and therefore lies on
the nine-point circle.
Moreover, since the lines T,T1' and T2T2 are parallel to the Euler line
and the lines T1T2 and TiT2 pass through L, we conclude that: a) the lines
T1T2 and TiT2 pass through M, and b) the centers of the two equilateral
pedal triangles lie on the line LM.
Problem 32. Given a quadrilateral ABCD, find the locus of points P such
that the radii of the circumcircles of the triangles ABP, BCP, CDP and
DAP are equal.
Problem 33. Let P be an arbitrary point on an equilateral hyperbola. Let
Q be the point symmetric to P with respect to the center of the hyperbola.
The circle with center P and radius PQ intersects the hyperbola at three
more points A, B and C. Prove that the triangle ABC is equilateral.
Problem 34. Let P be the center of an equilateral hyperbola passing
through the vertices of an inscribed quadrilateral ABCD. Prove that P
lies on the line connecting the center of the circumcircle and the centroid of
the quadrilateral ABCD.
Problem 35. Prove that points A, B, C, A', B', C' lie on a conic if and only
if there is a conic with respect to which both triangles ABC and A'B'C' are
self-polar.
Problem 36. A triangle ABC is self-polar with respect to a conic with
center O. Prove that this conic is homothetic to the conic passing through
the midpoints of the segments AB, BC, CA, OA, OB, OC.
FIGURE 4.7
Suppose now that a conic with center Q is tangent to the sides of the
triangle at points A', B', C', and let C1 be the intersection of the lines C'Q
and A'B'. Since Q is the center of the conic, the pole of the line C'Q is the
point at infinity of the line AB, and the pole of the line A'B' is C. Thus
the polar of C1 is the line cm, and therefore C1 lies on the median CM,.
By Theorem 3.9, the intersection of the lines CC' and MQ lies on
the midline Ma,Mb, i.e., Q is isotomically conjugate, with respect to the
triangle MaMbMC, to the image of the perspector P under the homothety
with center M and coefficient -.1. This immediately implies the assertion
of the theorem.
Theorem 4.7 implies that for each point, there is a unique inscribed conic
with center at that point. In particular, if the center of the conic coincides
with M, then M is also the perspector and the conic is the inscribed Steiner
ellipse, i.e., the preimage of the incircle under an affine transformation mak-
ing the triangle equilateral. Notice that the Steiner ellipse has the largest
area of all ellipses inscribed in a given triangle. This follows from the facts
that an equilateral triangle has the smallest area among all triangles cir-
cumscribed about a circle and that affine transformations preserve ratios of
areas.
Theorem 4.8. The center of an inscribed conic with perspector P is the
pole of the line PM with respect to the conic passing through A, B, C, M
and P (Figure 4.8).
Proof. This follows from the theorem just proved and Theorem 3.9.
Finally, consider the pencil of conics tangent to four given lines. Let U,
U' and V, V' be the foci of two conics from that pencil. Then the points of
each pair of foci are isogonally conjugate with respect to the triangle formed
by any three of the given lines. As was shown in 3.3, this implies that the
intersections of UV with U'V' and of U'V with UV' are also isogonally
conjugate with respect to all of these four triangles and therefore are the
4.2. INSCRIBED CONICS 107
FIGURE 4.8
foci of some conic from the pencil. Since the projections of the focus to the
given lines lie on a circle, it is not difficult to see that the locus of the foci
is a cubic, i.e., a curve of degree three. The correspondence between the
foci of each conic of the pencil gives rise to an involution on that cubic. We
have shown that for any two pairs of the corresponding points U, U' and
V, V' on the cubic, the intersection of the lines UV and U'V' also lies on
the cubic. As a limit case, we have that the tangents to the cubic at the
corresponding points U and U' intersect at the cubic and their intersection
point corresponds to the third intersection point of the cubic with the line
UU' (Figure 4.9).
Suppose a parabola is inscribed in a triangle. By Theorem 4.7, we have
that the point isotomically conjugate to the perspector of the parabola is a
point at infinity. If the triangle is equilateral, then the isotomic conjugation
coincides with the isogonal conjugation and the image of the line at infinity
is the circumcircle of the triangle. In the general case, we have the preimage
of that circle under an affine transformation making the triangle equilateral,
i.e., the circumscribed Steiner ellipse, whose tangents at the vertices of the
triangle are parallel to the opposite sides. Notice that this ellipse has the
smallest area among all ellipses circumscribed about the given triangle. Thus
we have proved the following.
Theorem 4.9. The locus of the perspectors of the parabolas inscribed in a
given triangle is the circumscribed Steiner ellipse.
Problem 37. A parabola is tangent to the sides of a triangle at points A',
B', and C'. Prove that the intersection of the line passing through C' and
parallel to the axis of the parabola with the line A'B' lies on the median
CMS.
Problem 38. Prove that the circumscribed and the inscribed Steiner el-
lipses are homothetic. Find the center and the coefficient of the homothety.
108 4. EUCLIDEAN PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 4.9
Problem 39. Prove that the locus of the centers of conics passing through
the vertices of a triangle and its centroid is the inscribed Steiner ellipse.
Problem 40. Suppose points P and P are isogonally conjugate with re-
spect to a triangle ABC, and let A', B', C' be the intersections of the sides of
the triangle and the lines connecting P with the centers of the corresponding
excircles. Prove that the lines AA', BY, CC' and PP' intersect at a single
point.
Problem 41. Suppose we are given a triangle and the center of a conic
inscribed in it. Determine if the conic is an ellipse or a hyperbola.
4.2. INSCRIBED CONICS 109
FIGURE 4.10
FIGURE 4.11
FIGURE 4.12
4.2. INSCRIBED CONICS 111
FIGURE 4.13
the sides of the triangle at points A2, B2 and C2. Then the midpoints of the
segments AlA2, B1B2 and C1C2 lie on a straight line.
FIGURE 4.14
Proof. Consider the parabola tangent to the sides of the triangle and the
line l1. By Theorems 1.11 and 1.7, the tangent to that parabola pass-
ing through H and different from l1 is perpendicular to 11, and therefore
it coincides with 12. By Theorem 1.10, the circumcircles of the triangles
A1A2H, B1B2H and C1C2H pass through the focus F of the parabola.
4.3. NORMALS TO CONICS. JOACHIMSTAHL'S CIRCLE 113
Hence the centers of those circles lie on the midpoint perpendicular to FH.
Since all these triangles are right, the centers of their circumcircles are the
midpoints of the hypotenuses, i.e., of the segments A1A2, B1B2 and C1C2
(Figure 4.14).1
When a point P moves along the circumcircle of the triangle, the mid-
point perpendiculars to PH are tangent to the conic with foci at H and
0 inscribed in the triangle (this follows from the construction described in
Theorem 3.4). Thus we have also proved that all those lines envelop the
conic with foci at 0 and H inscribed in the triangle.
We remark that the above theorem easily follows from Problem 18 at the
end of 3.1. For the centrally symmetric triangle we take the triangle whose
vertices are the reflections of the center of the circumcircle in the sides of the
triangle. Each line obtained this way gives rise to two perpendicular lines
passing through the orthocenter and generating the line from the Droz-
Farny theorem. The existence of those two lines can easily be proved by
computing the angles.
Proof. Given an arbitrary circle w with center Q, consider the locus of the
centers of the conics from the pencil generated by the circle and the conic
a. By Theorem 3.17, it is a conic and, since the pencil contains a circle, it
is an equilateral hyperbola (see Problem 29). We denote it y (Figure 4.15).
The infinite points of the hyperbola y are the fixed points of the involution
defined by the pencil on the line at infinity, i.e., the points of the axes of a.
Any point X on y is the center of a conic from the pencil. Its polar with
respect to the conic is the line at infinity, and therefore, by Theorem 3.15,
the polars of X with respect to all conics of the pencil are parallel. Since
the line QX is perpendicular to the polar of X with respect to the circle w,
it is also perpendicular to the polar of X with respect to the conic a. Hence
'The last part of the proof is a special case of Problem 7 at the end of Chapter 1.
114 4. EUCLIDEAN PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 4.15
FIGURE 4.16
4.4. THE PONCELET THEOREM FOR CONFOCAL ELLIPSES 115
Proof. Consider the case where the given conic is an ellipse. Let P' be the
point symmetric to P4 with respect to 0 (Figure 4.16). The assertion of the
theorem is equivalent to the statement that the conic of the centers of the
pencil determined by P1, P2, P3, P' is an equilateral hyperbola. Consider
the compression to the minor axis of the ellipse transforming it into a circle.
Since affine transformations take centers of conics into centers of conics, the
locus in question also transforms into a conic and, in fact, into an equilateral
hyperbola, since the new pencil contains a circle (the image of the original
ellipse). This hyperbola contains the images of 0 (i.e., 0 itself) and of the
midpoints of the segments P'P1, P'P2, P'P3. The homothety with respect to
the image of P' and with coefficient 2 transforms those points into the points
P4, P1, P2 and P3. But since there is only one equilateral hyperbola passing
through the images of P4, P1, P2 and P3, it must be the corresponding
Apollonius hyperbola (under the compression it is taken into an equilateral
hyperbola, because its asymptotes run in the directions of the axes of the
ellipse); i.e., the conic of the centers of the pencil determined by P1, P2, P3,
P is homothetic to the Apollonius hyperbola passing through P1, P2, P3,
P4, and is therefore an equilateral hyperbola. 0
The circle mentioned in the theorem is called Joachimstahl's circle.
Using complex affine transformations, one can also prove Theorem 4.15
for hyperbolas. In the case where the conic is a parabola, one of the four feet
of the normals passing through the given point is at infinity. Accordingly,
Theorem 4.15 is stated as follows:
The feet of the three normals to a parabola passing through a given point
and the vertex of the parabola lie on a circle (Figure 4.17).
FIGURE 4.17
the polar circles also form a pencil. Therefore the confocal conics form the
dual pencil.
We consider two confocal ellipses such that there is a polygon inscribed in
the larger ellipse and circumscribed about the smaller ellipse (Figure 4.18).
By the Poncelet theorem, there are infinitely many such polygons (if there
is at least one). It turns out that they have some interesting properties.
In 1.4 we established the following fact. Suppose a string is put on an
ellipse a which is then pulled tight using a pencil. If the pencil is rotated
about the ellipse, it will traverse another ellipse confocal with a.
As an important consequence of this fact we have the following theorem.
Theorem 4.16. If a convex n-gon is inscribed in a given ellipse a and has
the longest perimeter among all such n-gons, then it is circumscribed about
an ellipse an confocal with a.
FIGURE 4.18
FIGURE 4.19
Suppose this is not the case. Let a' be the ellipse with foci Mz_1 and
Mi+1 that passes through Mi (Figure 4.19). Then the tangent to a', which
is also the bisector of the exterior angle Mi_1MMMi+i, intersects a at some
other point M. But the sum Mi+IMx is larger than Mi_IMi +
Mi+1Mi, since M,, is outside the ellipse a'. Thus we have a convex n-gon
M1M2 ... Mi_iMMMi+l ... Mn whose perimeter is longer than the perimeter
of M1M2 ... Mn. But this contradicts our choice of the polygon.
4.4. THE PONCELET THEOREM FOR CONFOCAL ELLIPSES 117
Now we show that an ellipse can be inscribed in the n-gon M1M2 ... Mn.
Let F1 and F2 be the foci of the ellipse a. Consider the ellipse Kn that has
the same foci and is tangent to the line M1M2. The angle between the
second tangent to an passing through M2 and M2F1 equals LF1M2M1. But
such a line must be the line M2M3 since the angles M3M2M1 and F1M2F2
have equal exterior bisectors (the tangent to a at M2).
Similarly, examining the vertex M3, we see that an is tangent to M3M4,
etc.
By the Poncelet theorem, the polygon M1M2 ... Mn can be "rotated"
between a and an. We show that the perimeter of the polygon does not
change under this rotation.
Indeed, that perimeter can easily be computed. Let Qi be the tangency
points of an and the sides MiMi+1 By Theorem 1.6, for an arbitrary point
M on a, the quantity MX + MY+ - XY, where MX and MY are the
tangents to an, does not depend on M. Hence C = Qz_1Mi + Mi,Qi+ -
QiQi_1 does not depend on Mi. Computing the sum of the lengths of such
loops for each Mi, we have nC. Each side of the polygon will be counted
once, whereas the arcs of the ellipse are counted n - 1 times (the shortest of
the two arcs QiQi+i belongs to all loops except for the loop corresponding
to Mi+1). Thus the perimeter of the n-gon M1M2 ... Mn is equal to the
difference between nC and the perimeter of the ellipse an counted n - 1
times. Therefore it does not change under the rotation.
A convex Poncelet n-gon inscribed in one of the two confocal ellipses and
circumscribed about the other has yet another extremal property, which in
a sense is "dual" to the one established in Theorem 4.16.
Theorem 4.17. A convex n-gon circumscribed about a given ellipse a has
the shortest perimeter among all such n-gons if and only if all of its vertices
lie on an ellipse confocal with a.
Proof. Fix the tangency points Mi_1, Mi+1 of the sides of the polygon
and the ellipse. Let T be the intersection of the tangents at those points
(Figure 4.20). For the sake of definiteness, suppose that the arc Mi_1Mi+i
is less than half of the ellipse. We want to find a point Mi on it such
that the length of the polygonal line Mz_1XYMi+1, where X and Y are
the intersection points of the tangent passing through Mi with TMi_1
and TMi+1i is the least possible. Let X' and Y' be the tangency points
of the incircle of the triangle TXY and its sides TY and TX. Then
Mi_1XYMi+1 = Mz_1Y' + X'Mi+1i i.e., the desired minimum is attained
when the incircle is of largest possible radius. Therefore the incircle and the
ellipse are tangent to the line XY at the same point but on the opposite
sides. Thus, by Theorem 3.16, the center 0 of the ellipse lies on the line
IT', where I is the center of the incircle of the triangle TXY and T' is the
midpoint of XY.
Let F1 and F2 be the foci of the given ellipse. Since it is inscribed
in the triangle TXY, the lines F1X and F2X form equal angles with the
118 4. EUCLIDEAN PROPERTIES OF CURVES OF SECOND DEGREE
FIGURE 4.20
line IX, and the lines F1Y and F2Y form equal angles with the line IY.
Moreover, the line connecting the midpoints of the segments F1F2 and XY
passes through I. We now invoke Problem 26 (in 3.3). Since the angle XIY
is obtuse, the lengths of the polygonal lines F1XF2 and F1YF2 are equal,
i.e., X and Y lie on an ellipse confocal with K. This, obviously, implies the
assertion of the theorem.
The case where the arc MZ_IMZ+1 is longer than half of the ellipse is
argued similarly except that the incircle of the triangle TXY must be re-
placed by an excircle. Finally, if Mz_1 and Mi+1 are antipodal points, then
the desired assertion is established by passing to limit or by deducing it from
the fact that the sum of the sides of a parallelogram circumscribed about an
ellipse cannot be less than the sum of its axes. This last assertion, in turn,
can be proved by a simple calculation.
Problem 42. Suppose that an ellipse and a circle are externally tangent
to each other and that their common tangents are parallel. Prove that the
distance between their centers equals the sum of the semiaxes of the ellipse.
Chapter 5
4 4
Let us find its foci. Suppose a point X moves along the hyperbola
toward infinity. Then the lines F1X and F2X will tend to becoming parallel
(where F1 and F2 are the foci of the hyperbola), and therefore the quantity
JF1X - F2X I equals the length of the projection of the segment F1 F2 to the
axis Ox. On the other hand, it also equals the real axis of the hyperbola,
i.e., 2i. Since the angle between F1F2 and Ox equals 45°, the quantity
FIF2 equals 2V2-. = 4. Hence OFI = OF2 = 2 and therefore F1 has
coordinates and F2 has coordinates (-/, -/).
2. Connect the tangency points and the vertices of the polygon with F.
Color the obtained angles containing black sides, red and containing white
sides, blue. By Theorem 1.4, the angles at the same vertex of the polygon
are equal and have different colors. Hence the sum of the red angles equals
the sum of the blue angles, i.e., 180°.
3. The assumption implies that the lines symmetric to the diagonal AC
with respect to the bisectors of the angles A and C intersect at a single
point P of the diagonal BD. Applying the sine theorem to triangles ABP,
ADP, ABL and ADL, where L is the intersection of the diagonals, we have
BL BP BL BP
- AD . Similarly, = D Therefore AD = LEI,, whence the
desired assertion.
4. The eccentricity of all equilateral hyperbolas obviously equals
(check this!). Hence the directrices of the hyperbolas in question are at a
distance FP// away from P. Therefore they are tangent to the circle w
with center P and radius FP/V12-. It is easy to see that they envelop the
entire circle (because each tangent to the circle gives rise to an equilateral
119
120 5. SOLUTIONS TO THE PROBLEMS
hyperbola with focus P and directrix the tangent). On the other hand, the
directrices are taken into asymptotes under a rotational homothety centered
at F (with rotation through ±45° and the coefficient of homothety 1/V/-2).
Hence all asymptotes of such equilateral hyperbolas will be tangent to one
of the two circles obtained from w by the above homothety (see Figure 5.1).
FIGURE 5.1
FIGURE 5.2
6. Let X and Y denote the positions of the travelers and A the intersec-
tion of the roads. The intersection point of the midpoint perpendiculars to
AX and AY moves along some line 1 (since its projections to the roads move
with constant speed). At all times, the circumcircle of the triangle AXY
5. SOLUTIONS TO THE PROBLEMS 121
FIGURE 5.3
FIGURE 5.4
14. Assume first that the conic intersecting the triangle at the points
in question is a circle. Then, for example, BA1 BA2 = BC1 BC2. Hence
this ratio equals 1. It now immediately follows that if the conic is an ellipse,
then this quantity is also preserved, since an affine transformation taking
that ellipse into a circle does not change the ratio ("we multiply and divide
by two collinear segments"). This argument does not work for a hyperbola,
since there is no affine transformation taking a circle to a hyperbola. But
this can be done by a projective transformation.
5. SOLUTIONS TO THE PROBLEMS 123
FIGURE 5.5
the conic. Notice that our proof does now use any specific order of the
points.
Based on this we can now prove the Poncelet theorem for triangles.
Suppose a triangle ABC is inscribed in a conic al and circumscribed about
a conic a2. Take an arbitrary point D on al which is outside a2 and draw
tangents to a2. Suppose they intersect al again at points E and F. We
need to show that EF is tangent to a2. As we have shown, there is a conic
tangent to the lines AB, BC, CA, DE, EF and FD. On the other hand,
there is a unique conic tangent to the five lines AB, BC, CA, DE, DF, and
this is a2. Therefore EF is also tangent to a2, which is the desired claim.
17. Hint. The quadrilateral formed by the given lines and the asymp-
totes is circumscribed about the hyperbola.
18. By the converse of Brianchon's theorem, there is a conic tangent to
the sides of both triangles. Suppose an arbitrary line tangent to that conic
intersects AC in P and BC in Q. It suffices to show that the lines PB' and
QA' are parallel. But this follows from Brianchon's theorem applied to the
hexagon A'B'XQPY, circumscribed about the same conic, where X and Y
are the infinite points of the lines BC and AC.
19. Let X and Y be two points on the line at infinity whose correspond-
ing directions are perpendicular. Draw the lines: through A and B, parallel
to the direction of X, and through C and H, parallel to the direction of Y.
Let UV be the diagonal formed by those lines of the rectangle and B' the
foot of the triangle's height dropped from B (Figure 5.6).
Since the quadrilaterals BB'CV and AUB'H are inscribed in the circles
with diameters BC and AH, we have LAB'U = LAHU and LVBC =
LVBC. But LAHU = LVBC as the angles with perpendicular sides, and
therefore U, B', V lie on a straight line. By the converse of Pascal's theorem,
the hexagon AXBHYC is inscribed in a conic, i.e., the equilateral hyperbola
ABCXY passes through H.
Conversely, suppose the conic passes through A, B, C and H. Since A,
B, C, H are not the vertices of a convex quadrilateral, that conic must be a
hyperbola. If X is one of the points of its intersection with the line at infinity,
5. SOLUTIONS TO THE PROBLEMS 125
FIGURE 5.6
with the center 0 of the hyperbola is a median of the right triangle OXY.
Hence it forms the same angles with the asymptotes of the hyperbola as the
line AB.
2. Let P be an intersection point of two equilateral hyperbolas with
center O. Since the tangents to the hyperbolas at that point are symmetric
to OP with respect to lines parallel to the asymptotes, the angle between
them is twice as large as the angle between the asymptotes.
23. Apply the polar transformation with respect to a circle centered at
the center 0 of the ellipse. Then the sides of a rhombus will transform into
points from which the image of the ellipse (which is also an ellipse) is seen at
a right angle. By Theorem 1.5, this is a circle also centered at O. Therefore
its image, i.e., the envelope of all the rhombi, is a circle with center O.
24. The circle and the hyperbola form a double tangent pencil. Under
the polar correspondence with respect to the circle the hyperbola transforms
into a conic from that pencil. The points at infinity of that conic are the im-
ages of the asymptotes of the hyperbola, which, obviously, coincide with the
points at infinity of the hyperbola. Therefore the hyperbola is transformed
into itself.
We shall now show that the circle is self-polar, too, with respect to the
hyperbola. Choose two perpendicular lines in space and consider the two
cones of revolution obtained by rotating the lines about the bisectors of the
angles between them. Those lines are common rulings of the cones along
which the cones are tangent to each other. Given a plane perpendicular to
the axis of one of the cones, the cross-section of that cone is a circle and the
cross-section of the other is an equilateral hyperbola tangent to the circle
at its vertices (Figure 5.7). Therefore there is a projective transformation
interchanging the circle and the hyperbola. Moreover, it preserves self-
polarity.
Notice also that if a secant plane is perpendicular to the common ruling
of the cones, then the cross-sections are equal parabolas tangent to each
other at their vertices. Hence such parabolas are also self-polar with respect
to each other.
25. Since T lies on the Gauss line, it is the center of a conic inscribed
in the quadrilateral. Suppose the quadrilateral is not circumscribed; then
that conic is not a circle, i.e., the distance from its center to the tangent
takes any value at most four times. By assumption, the opposite sides of
the quadrilateral are equidistant from the center. Hence either at least one
pair consists of parallel sides and the quadrilateral is a trapezoid, or the
sides in each pair are symmetric with respect to one of the axes of the conic.
Those axes cannot coincide, because otherwise a circle could be inscribed
in the quadrilateral. Hence the bisectors of the angles between the opposite
sides of the quadrilateral are perpendicular, which is equivalent to its being
inscribed. The converse for an inscribed quadrilateral is proved similarly,
and for a circumscribed quadrilateral and a trapezoid it is obvious.
5. SOLUTIONS TO THE PROBLEMS 127
FIGURE 5.7
37. Hint. Brianchon's theorem implies that the diagonals of the quadri-
lateral circumscribed about a conic and the lines connecting the tangency
points of the conic with its opposite sides intersect at a single point. Apply-
ing this observation to the quadrilateral formed by the sides of the triangle
ABC and the line at infinity, we have the desired assertion.
38. Answer. The center of the homothety is the centroid of the triangle,
and the coefficient is 2.
39. The preceding problem implies that the midpoints of the segments
connecting the centroid with the vertices lie on the Steiner ellipse. Therefore
that ellipse has six common points with the conic of the centers (that this is
indeed a conic follows from Theorem 3.17). Those points are the midpoints
of the sides and of the segments.
40. Let Ia,, Ib and I, be the centers of the excircles of the triangle
ABC. Since ABC is the orthotriangle of the triangle IaIbIc, the polars of
P with respect to the equilateral hyperbolas passing through Ia, Ib and Ic
pass through P'. By Theorem 3.9, the lines AA', BY and CC' intersect at
a single point, which is the pole of the line PI, where I is the center of the
incircle of the triangle ABC, with respect to the conic IalbIciP. But that
conic is an equilateral hyperbola so that the obtained point lies on PP'.
41. Any point in the plane can be the perspector of a conic inscribed
in the given triangle, and such a conic is unique. It is clear that the conic
changes continuously as the perspector moves. Therefore the perspectors of
ellipses lie inside the circumscribed Steiner ellipse, whereas the perspectors
of hyperbolas lie outside. Now apply Theorem 4.7. Under the isotomic
conjugation, the interior of the triangle will transform into itself, and the
points of the segment bounded, for example, by the side AB of the triangle
and the arc of the ellipse subtended by it and not containing the third
vertex C will transform into the points of the angle vertical to the angle C.
Accordingly, the set of centers of the inscribed ellipses is the interior of the
midline triangle and the three angles vertical to its angles.
42. Let X and Y be the tangency points of the ellipse and the parallel
tangents, T the tangency point of the ellipse and the circle, U and V the
intersection points of the tangent to the ellipse at T and the parallel tangents,
and U' and V the intersection points of the parallel tangents and some other
tangent to the ellipse. Arguing as in the proof of the last theorem, we see
that XU + UV + VY < X U' + U'V' + V'Y, i.e., U and V lie on an ellipse
confocal with the given ellipse. Moreover, OU and OV, where 0 is the
center of the circle, are tangent to that ellipse and LUOV = 90°. Therefore
0 lies on the circle centered at the center of the ellipse whose radius equals
the sum of the semiaxes of the ellipse.
Bibliography
[1] P. S. Aleksandrov, Lectures on analytic geometry. Nauka, Moscow, 1968. (Russian)
[2] M. Berger, Geometry. I. Springer-Verlag, Berlin, 1994.
[3] , Geometry. II. Springer-Verlag, Berlin, 1987.
[4] H. S. M. Coxeter, The real projective plane. Springer-Verlag, New York, 1993.
[5] H. S. M. Coxeter and S. L. Greitzer, Geometry revisited. MAA, New York, 1967.
[6] J.-P. Ehrmann and F. van Lamoen, A projective generalization of the Droz-Farny line
theorem. Forum Geom. 4 (2004), 225-227 (electronic).
[7] L. A. Emelyanov and T. L. Emelyanova, Feuerbach's family. Matematicheskoe
prosveshchenie, Series 3, 6 (2004), 78-92. (Russian)
[8] D. Hilbert and S. Cohn-Vossen, Geometry and the imagination. Chelsea Publishing
Company, New York, Amer. Math. Soc., Providence, RI, 1952.
[9] J. Lemaire, L'hyperbole equilatere. Vuibert, Paris.
[10] I. M. Yaglom, Geometric transformations. Random House, New York, 1962.
[11] , Geometric transformations. II. Random House, New York, 1968.
[12] A. A. Zaslavsky, Geometric transformations. MCCME, Moscow, 2003. (Russian)
131
Index
isotomic conjugation, 88 ellipse, 1
Brocard, 48
asymptotes of a hyperbola, 2 Steiner
axis circumscribed, 53, 107
of a hyperbola inscribed, 52
imaginary, 2
real, 2 hyperbola, 2
of a parabola, 2 Kiepert, 104
of an ellipse Apollonius, 114
major, 1 equilateral, 2, 99
minor, 1 Feuerbach, 103
radical, 57
inversion, 27
Brocard angle, 49 involution, 93
isogonal conjugate, 12
circle isogonal conjugation, 41, 87
Apollonius, 41 isotomic conjugation, 56
Ceva, 39
Euler, 28 lemma
Fermat-Apollonius, 13 Simson's, 22
Joachimstahl's, 115 line
nine-point, 28 Aubert, 109
pedal, 38 Gauss, 88, 109
coaxial circles, 58 Simson's, 23
complete quadrilateral, 109
conic, 2, 17 normal, 113
conic section, 17
correspondence parabola, 2, 19
polar, 68 pencil, 78
cross-ratio, 31, 32 of circles, 58
curve of second degree, 5 elliptic, 58
curve of second degree, 2 hyperbolic, 58
parabolic, 58
Dandelin spheres, 17 perspector, 105
directrix, 18 point
duality principle, 36 Apollonius, 53
Brocard, 48
eccentricity, 18 Feuerbach, 29
133
134 INDEX